Merge pull request #916 from pimoroni/examples/audio

i2s audio examples
pull/946/head
Philip Howard 2024-06-03 11:35:59 +01:00 zatwierdzone przez GitHub
commit bf6fa4bc45
Nie znaleziono w bazie danych klucza dla tego podpisu
ID klucza GPG: B5690EEEBB952194
33 zmienionych plików z 1646 dodań i 0 usunięć

Plik binarny nie jest wyświetlany.

Plik binarny nie jest wyświetlany.

Plik binarny nie jest wyświetlany.

Wyświetl plik

@ -0,0 +1,276 @@
# SPDX-FileCopyrightText: 2023 Christopher Parrott for Pimoroni Ltd
#
# SPDX-License-Identifier: MIT
import os
import math
import struct
from machine import I2S, Pin
"""
A class for playing Wav files out of an I2S audio amp. It can also play pure tones.
This code is based heavily on the work of Mike Teachman, at:
https://github.com/miketeachman/micropython-i2s-examples/blob/master/examples/wavplayer.py
"""
class WavPlayer:
# Internal states
PLAY = 0
PAUSE = 1
FLUSH = 2
STOP = 3
NONE = 4
MODE_WAV = 0
MODE_TONE = 1
# Default buffer length
SILENCE_BUFFER_LENGTH = 1000
WAV_BUFFER_LENGTH = 10000
INTERNAL_BUFFER_LENGTH = 20000
TONE_SAMPLE_RATE = 44_100
TONE_BITS_PER_SAMPLE = 16
TONE_FULL_WAVES = 2
def __init__(self, id, sck_pin, ws_pin, sd_pin, amp_enable=None, ibuf_len=INTERNAL_BUFFER_LENGTH, root="/"):
self.__id = id
self.__sck_pin = sck_pin
self.__ws_pin = ws_pin
self.__sd_pin = sd_pin
self.__ibuf_len = ibuf_len
self.__enable = None
if amp_enable is not None:
self.__enable = Pin(amp_enable, Pin.OUT)
# Set the directory to search for files in
self.set_root(root)
self.__state = WavPlayer.NONE
self.__mode = WavPlayer.MODE_WAV
self.__wav_file = None
self.__loop_wav = False
self.__first_sample_offset = None
self.__flush_count = 0
self.__audio_out = None
# Allocate a small array of blank audio samples used for silence
self.__silence_samples = bytearray(self.SILENCE_BUFFER_LENGTH)
# Allocate a larger array for WAV audio samples, using a memoryview for more efficient access
self.__wav_samples_mv = memoryview(bytearray(self.WAV_BUFFER_LENGTH))
# Reserve a variable for audio samples used for tones
self.__tone_samples = None
self.__queued_samples = None
def set_root(self, root):
self.__root = root.rstrip("/") + "/"
def play_wav(self, wav_file, loop=False):
if os.listdir(self.__root).count(wav_file) == 0:
raise ValueError(f"'{wav_file}' not found")
self.__stop_i2s() # Stop any active playback and terminate the I2S instance
self.__wav_file = open(self.__root + wav_file, "rb") # Open the chosen WAV file in read-only, binary mode
self.__loop_wav = loop # Record if the user wants the file to loop
# Parse the WAV file, returning the necessary parameters to initialise I2S communication
format, sample_rate, bits_per_sample, self.__first_sample_offset, self.sample_size = WavPlayer.__parse_wav(self.__wav_file)
# Keep a track of total bytes read from WAV File
self.total_bytes_read = 0
self.__wav_file.seek(self.__first_sample_offset) # Advance to first byte of sample data
self.__start_i2s(bits=bits_per_sample,
format=format,
rate=sample_rate,
state=WavPlayer.PLAY,
mode=WavPlayer.MODE_WAV)
def play_tone(self, frequency, amplitude):
if frequency < 20.0 or frequency > 20_000:
raise ValueError("frequency out of range. Expected between 20Hz and 20KHz")
if amplitude < 0.0 or amplitude > 1.0:
raise ValueError("amplitude out of range. Expected 0.0 to 1.0")
# Create a buffer containing the pure tone samples
samples_per_cycle = self.TONE_SAMPLE_RATE // frequency
sample_size_in_bytes = self.TONE_BITS_PER_SAMPLE // 8
samples = bytearray(self.TONE_FULL_WAVES * samples_per_cycle * sample_size_in_bytes)
range = pow(2, self.TONE_BITS_PER_SAMPLE) // 2
format = "<h" if self.TONE_BITS_PER_SAMPLE == 16 else "<l"
# Populate the buffer with multiple cycles to avoid it completing too quickly and causing drop outs
for i in range(samples_per_cycle * self.TONE_FULL_WAVES):
sample = int((range - 1) * (math.sin(2 * math.pi * i / samples_per_cycle)) * amplitude)
struct.pack_into(format, samples, i * sample_size_in_bytes, sample)
# Are we not already playing tones?
if not (self.__mode == WavPlayer.MODE_TONE and (self.__state == WavPlayer.PLAY or self.__state == WavPlayer.PAUSE)):
self.__stop_i2s() # Stop any active playback and terminate the I2S instance
self.__tone_samples = samples
self.__start_i2s(bits=self.TONE_BITS_PER_SAMPLE,
format=I2S.MONO,
rate=self.TONE_SAMPLE_RATE,
state=WavPlayer.PLAY,
mode=WavPlayer.MODE_TONE)
else:
self.__queued_samples = samples
self.__state = WavPlayer.PLAY
def pause(self):
if self.__state == WavPlayer.PLAY:
self.__state = WavPlayer.PAUSE # Enter the pause state on the next callback
def resume(self):
if self.__state == WavPlayer.PAUSE:
self.__state = WavPlayer.PLAY # Enter the play state on the next callback
def stop(self):
if self.__state == WavPlayer.PLAY or self.__state == WavPlayer.PAUSE:
if self.__mode == WavPlayer.MODE_WAV:
# Enter the flush state on the next callback and close the file
# It is done in this order to prevent the callback entering the play
# state after we close the file but before we change the state)
self.__state = WavPlayer.FLUSH
self.__wav_file.close()
else:
self.__state = WavPlayer.STOP
def is_playing(self):
return self.__state != WavPlayer.NONE and self.__state != WavPlayer.STOP
def is_paused(self):
return self.__state == WavPlayer.PAUSE
def __start_i2s(self, bits=16, format=I2S.MONO, rate=44_100, state=STOP, mode=MODE_WAV):
import gc
gc.collect()
self.__audio_out = I2S(
self.__id,
sck=self.__sck_pin,
ws=self.__ws_pin,
sd=self.__sd_pin,
mode=I2S.TX,
bits=bits,
format=format,
rate=rate,
ibuf=self.__ibuf_len,
)
self.__state = state
self.__mode = mode
self.__flush_count = self.__ibuf_len // self.SILENCE_BUFFER_LENGTH + 1
self.__audio_out.irq(self.__i2s_callback)
self.__audio_out.write(self.__silence_samples)
if self.__enable is not None:
self.__enable.on()
def __stop_i2s(self):
self.stop() # Stop any active playback
while self.is_playing(): # and wait for it to complete
pass
if self.__enable is not None:
self.__enable.off()
if self.__audio_out is not None:
self.__audio_out.deinit() # Deinit any active I2S comms
self.__state == WavPlayer.NONE # Return to the none state
def __i2s_callback(self, arg):
# PLAY
if self.__state == WavPlayer.PLAY:
if self.__mode == WavPlayer.MODE_WAV:
num_read = self.__wav_file.readinto(self.__wav_samples_mv) # Read the next section of the WAV file
self.total_bytes_read += num_read
# Have we reached the end of the file?
if num_read == 0:
# Do we want to loop the WAV playback?
if self.__loop_wav:
_ = self.__wav_file.seek(self.__first_sample_offset) # Play again, so advance to first byte of sample data
else:
self.__wav_file.close() # Stop playing, so close the file
self.__state = WavPlayer.FLUSH # and enter the flush state on the next callback
self.__audio_out.write(self.__silence_samples) # In both cases play silence to end this callback
else:
if num_read > 0 and num_read < self.WAV_BUFFER_LENGTH:
num_read = num_read - (self.total_bytes_read - self.sample_size)
self.__audio_out.write(self.__wav_samples_mv[: num_read]) # We are within the file, so write out the next audio samples
else:
if self.__queued_samples is not None:
self.__tone_samples = self.__queued_samples
self.__queued_samples = None
self.__audio_out.write(self.__tone_samples)
# PAUSE or STOP
elif self.__state == WavPlayer.PAUSE or self.__state == WavPlayer.STOP:
self.__audio_out.write(self.__silence_samples) # Play silence
# FLUSH
elif self.__state == WavPlayer.FLUSH:
# Flush is used to allow the residual audio samples in the internal buffer to be written
# to the I2S peripheral. This step avoids part of the sound file from being cut off
if self.__flush_count > 0:
self.__flush_count -= 1
else:
self.__state = WavPlayer.STOP # Enter the stop state on the next callback
self.__audio_out.write(self.__silence_samples) # Play silence
# NONE
elif self.__state == WavPlayer.NONE:
pass
@staticmethod
def __parse_wav(wav_file):
chunk_ID = wav_file.read(4)
if chunk_ID != b"RIFF":
raise ValueError("WAV chunk ID invalid")
_ = wav_file.read(4) # chunk_size
format = wav_file.read(4)
if format != b"WAVE":
raise ValueError("WAV format invalid")
sub_chunk1_ID = wav_file.read(4)
if sub_chunk1_ID != b"fmt ":
raise ValueError("WAV sub chunk 1 ID invalid")
_ = wav_file.read(4) # sub_chunk1_size
_ = struct.unpack("<H", wav_file.read(2))[0] # audio_format
num_channels = struct.unpack("<H", wav_file.read(2))[0]
if num_channels == 1:
format = I2S.MONO
else:
format = I2S.STEREO
sample_rate = struct.unpack("<I", wav_file.read(4))[0]
# if sample_rate != 44_100 and sample_rate != 48_000:
# raise ValueError(f"WAV sample rate of {sample_rate} invalid. Only 44.1KHz or 48KHz audio are supported")
_ = struct.unpack("<I", wav_file.read(4))[0] # byte_rate
_ = struct.unpack("<H", wav_file.read(2))[0] # block_align
bits_per_sample = struct.unpack("<H", wav_file.read(2))[0]
# usually the sub chunk2 ID ("data") comes next, but
# some online MP3->WAV converters add
# binary data before "data". So, read a fairly large
# block of bytes and search for "data".
binary_block = wav_file.read(200)
offset = binary_block.find(b"data")
if offset == -1:
raise ValueError("WAV sub chunk 2 ID not found")
wav_file.seek(40)
sub_chunk2_size = struct.unpack("<I", wav_file.read(4))[0]
return (format, sample_rate, bits_per_sample, 44 + offset, sub_chunk2_size)

Plik binarny nie jest wyświetlany.

Plik binarny nie jest wyświetlany.

Wyświetl plik

@ -0,0 +1,91 @@
from machine import Timer
from audio import WavPlayer
from cosmic import CosmicUnicorn
from picographics import PicoGraphics, DISPLAY_COSMIC_UNICORN as DISPLAY
import time
cu = CosmicUnicorn()
graphics = PicoGraphics(DISPLAY)
graphics.set_font("bitmap6")
WHITE = graphics.create_pen(255, 255, 255)
BLUE = graphics.create_pen(0, 0, 255)
CLEAR = graphics.create_pen(0, 0, 0)
RED = graphics.create_pen(255, 0, 0)
GREEN = graphics.create_pen(0, 255, 0)
cu.set_brightness(0.7)
audio = WavPlayer(0, 10, 11, 9, amp_enable=22)
class Countdown(object):
def __init__(self):
self.timer_running = False
self.total_seconds = 0
self.timer = None
def process_input(self):
if cu.is_pressed(CosmicUnicorn.SWITCH_VOLUME_UP):
self.total_seconds += 1
if cu.is_pressed(CosmicUnicorn.SWITCH_VOLUME_DOWN):
if self.total_seconds > 0:
self.total_seconds -= 1
if cu.is_pressed(CosmicUnicorn.SWITCH_SLEEP):
self.start_timer()
def display_time(self):
seconds = self.total_seconds % (24 * 3600)
seconds %= 3600
minutes = seconds // 60
seconds %= 60
# Add leading zeros to the minutes and seconds
if len(str(minutes)) == 1:
minutes = "0{}".format(minutes)
if len(str(seconds)) == 1:
seconds = "0{}".format(seconds)
return "{}:{}".format(minutes, seconds)
def draw(self):
graphics.set_pen(graphics.create_pen(0, 0, 0))
graphics.clear()
graphics.set_pen(BLUE)
graphics.circle(0, 0, 12)
graphics.set_pen(GREEN)
graphics.circle(25, 30, 5)
graphics.set_pen(RED)
graphics.circle(0, 32, 12)
graphics.set_pen(CLEAR)
graphics.rectangle(0, 11, CosmicUnicorn.WIDTH, 9)
graphics.set_pen(WHITE)
graphics.text(self.display_time(), 4, 12, -1, 1)
cu.update(graphics)
def start_timer(self):
if not self.timer_running:
self.timer = Timer(mode=Timer.PERIODIC, period=1000, callback=self.countdown)
self.timer_running = True
def reset(self):
self.timer.deinit()
self.timer_running = False
def countdown(self, arg):
if self.total_seconds == 0:
audio.play_wav("doorbell.wav", False)
self.reset()
else:
self.total_seconds -= 1
count = Countdown()
while 1:
count.process_input()
count.draw()
time.sleep(0.07)

Plik binarny nie jest wyświetlany.

Wyświetl plik

@ -0,0 +1,74 @@
from audio import WavPlayer
from cosmic import CosmicUnicorn
from picographics import PicoGraphics, DISPLAY_COSMIC_UNICORN as DISPLAY
from time import sleep
cu = CosmicUnicorn()
graphics = PicoGraphics(DISPLAY)
audio = WavPlayer(0, 10, 11, 9, amp_enable=22)
WHITE = graphics.create_pen(255, 255, 255)
RED = graphics.create_pen(255, 0, 0)
class Menu(object):
def __init__(self):
self.items = ["Pew 1", "Pew 2", "Pew 3"]
self.selected = 0
# A function to draw only the menu elements.
# Helps to keep our main function tidy!
def draw_menu(self):
graphics.set_pen(WHITE)
for item in range(len(self.items)):
if self.selected == item:
graphics.set_pen(RED)
graphics.text(self.items[item], 0, 2 + item * 10, 31, 1)
graphics.set_pen(WHITE)
# Make changes based on the currently selected menu item
def process_selected(self):
if self.selected == 0:
audio.play_wav("Pew1.wav", False)
if self.selected == 1:
audio.play_wav("Pew2.wav", False)
if self.selected == 2:
audio.play_wav("Pew3.wav", False)
menu = Menu()
graphics.set_font("bitmap6")
cu.set_brightness(0.7)
while True:
graphics.set_pen(graphics.create_pen(0, 0, 0))
graphics.clear()
if cu.is_pressed(CosmicUnicorn.SWITCH_BRIGHTNESS_DOWN):
audio.play_wav("buttonbeep.wav", False)
if menu.selected + 1 < len(menu.items):
menu.selected += 1
else:
menu.selected = 0
if cu.is_pressed(CosmicUnicorn.SWITCH_BRIGHTNESS_UP):
audio.play_wav("buttonbeep.wav", False)
if menu.selected > 0:
menu.selected -= 1
else:
menu.selected = len(menu.items) - 1
if cu.is_pressed(CosmicUnicorn.SWITCH_SLEEP):
menu.process_selected()
menu.draw_menu()
cu.update(graphics)
sleep(0.2)

Wyświetl plik

@ -0,0 +1,8 @@
from audio import WavPlayer
sound = WavPlayer(0, 10, 11, 9, amp_enable=22)
sound.play_wav("beepboop.wav", False)
while sound.is_playing():
pass

Plik binarny nie jest wyświetlany.

Plik binarny nie jest wyświetlany.

Plik binarny nie jest wyświetlany.

Wyświetl plik

@ -0,0 +1,276 @@
# SPDX-FileCopyrightText: 2023 Christopher Parrott for Pimoroni Ltd
#
# SPDX-License-Identifier: MIT
import os
import math
import struct
from machine import I2S, Pin
"""
A class for playing Wav files out of an I2S audio amp. It can also play pure tones.
This code is based heavily on the work of Mike Teachman, at:
https://github.com/miketeachman/micropython-i2s-examples/blob/master/examples/wavplayer.py
"""
class WavPlayer:
# Internal states
PLAY = 0
PAUSE = 1
FLUSH = 2
STOP = 3
NONE = 4
MODE_WAV = 0
MODE_TONE = 1
# Default buffer length
SILENCE_BUFFER_LENGTH = 1000
WAV_BUFFER_LENGTH = 10000
INTERNAL_BUFFER_LENGTH = 20000
TONE_SAMPLE_RATE = 44_100
TONE_BITS_PER_SAMPLE = 16
TONE_FULL_WAVES = 2
def __init__(self, id, sck_pin, ws_pin, sd_pin, amp_enable=None, ibuf_len=INTERNAL_BUFFER_LENGTH, root="/"):
self.__id = id
self.__sck_pin = sck_pin
self.__ws_pin = ws_pin
self.__sd_pin = sd_pin
self.__ibuf_len = ibuf_len
self.__enable = None
if amp_enable is not None:
self.__enable = Pin(amp_enable, Pin.OUT)
# Set the directory to search for files in
self.set_root(root)
self.__state = WavPlayer.NONE
self.__mode = WavPlayer.MODE_WAV
self.__wav_file = None
self.__loop_wav = False
self.__first_sample_offset = None
self.__flush_count = 0
self.__audio_out = None
# Allocate a small array of blank audio samples used for silence
self.__silence_samples = bytearray(self.SILENCE_BUFFER_LENGTH)
# Allocate a larger array for WAV audio samples, using a memoryview for more efficient access
self.__wav_samples_mv = memoryview(bytearray(self.WAV_BUFFER_LENGTH))
# Reserve a variable for audio samples used for tones
self.__tone_samples = None
self.__queued_samples = None
def set_root(self, root):
self.__root = root.rstrip("/") + "/"
def play_wav(self, wav_file, loop=False):
if os.listdir(self.__root).count(wav_file) == 0:
raise ValueError(f"'{wav_file}' not found")
self.__stop_i2s() # Stop any active playback and terminate the I2S instance
self.__wav_file = open(self.__root + wav_file, "rb") # Open the chosen WAV file in read-only, binary mode
self.__loop_wav = loop # Record if the user wants the file to loop
# Parse the WAV file, returning the necessary parameters to initialise I2S communication
format, sample_rate, bits_per_sample, self.__first_sample_offset, self.sample_size = WavPlayer.__parse_wav(self.__wav_file)
# Keep a track of total bytes read from WAV File
self.total_bytes_read = 0
self.__wav_file.seek(self.__first_sample_offset) # Advance to first byte of sample data
self.__start_i2s(bits=bits_per_sample,
format=format,
rate=sample_rate,
state=WavPlayer.PLAY,
mode=WavPlayer.MODE_WAV)
def play_tone(self, frequency, amplitude):
if frequency < 20.0 or frequency > 20_000:
raise ValueError("frequency out of range. Expected between 20Hz and 20KHz")
if amplitude < 0.0 or amplitude > 1.0:
raise ValueError("amplitude out of range. Expected 0.0 to 1.0")
# Create a buffer containing the pure tone samples
samples_per_cycle = self.TONE_SAMPLE_RATE // frequency
sample_size_in_bytes = self.TONE_BITS_PER_SAMPLE // 8
samples = bytearray(self.TONE_FULL_WAVES * samples_per_cycle * sample_size_in_bytes)
range = pow(2, self.TONE_BITS_PER_SAMPLE) // 2
format = "<h" if self.TONE_BITS_PER_SAMPLE == 16 else "<l"
# Populate the buffer with multiple cycles to avoid it completing too quickly and causing drop outs
for i in range(samples_per_cycle * self.TONE_FULL_WAVES):
sample = int((range - 1) * (math.sin(2 * math.pi * i / samples_per_cycle)) * amplitude)
struct.pack_into(format, samples, i * sample_size_in_bytes, sample)
# Are we not already playing tones?
if not (self.__mode == WavPlayer.MODE_TONE and (self.__state == WavPlayer.PLAY or self.__state == WavPlayer.PAUSE)):
self.__stop_i2s() # Stop any active playback and terminate the I2S instance
self.__tone_samples = samples
self.__start_i2s(bits=self.TONE_BITS_PER_SAMPLE,
format=I2S.MONO,
rate=self.TONE_SAMPLE_RATE,
state=WavPlayer.PLAY,
mode=WavPlayer.MODE_TONE)
else:
self.__queued_samples = samples
self.__state = WavPlayer.PLAY
def pause(self):
if self.__state == WavPlayer.PLAY:
self.__state = WavPlayer.PAUSE # Enter the pause state on the next callback
def resume(self):
if self.__state == WavPlayer.PAUSE:
self.__state = WavPlayer.PLAY # Enter the play state on the next callback
def stop(self):
if self.__state == WavPlayer.PLAY or self.__state == WavPlayer.PAUSE:
if self.__mode == WavPlayer.MODE_WAV:
# Enter the flush state on the next callback and close the file
# It is done in this order to prevent the callback entering the play
# state after we close the file but before we change the state)
self.__state = WavPlayer.FLUSH
self.__wav_file.close()
else:
self.__state = WavPlayer.STOP
def is_playing(self):
return self.__state != WavPlayer.NONE and self.__state != WavPlayer.STOP
def is_paused(self):
return self.__state == WavPlayer.PAUSE
def __start_i2s(self, bits=16, format=I2S.MONO, rate=44_100, state=STOP, mode=MODE_WAV):
import gc
gc.collect()
self.__audio_out = I2S(
self.__id,
sck=self.__sck_pin,
ws=self.__ws_pin,
sd=self.__sd_pin,
mode=I2S.TX,
bits=bits,
format=format,
rate=rate,
ibuf=self.__ibuf_len,
)
self.__state = state
self.__mode = mode
self.__flush_count = self.__ibuf_len // self.SILENCE_BUFFER_LENGTH + 1
self.__audio_out.irq(self.__i2s_callback)
self.__audio_out.write(self.__silence_samples)
if self.__enable is not None:
self.__enable.on()
def __stop_i2s(self):
self.stop() # Stop any active playback
while self.is_playing(): # and wait for it to complete
pass
if self.__enable is not None:
self.__enable.off()
if self.__audio_out is not None:
self.__audio_out.deinit() # Deinit any active I2S comms
self.__state == WavPlayer.NONE # Return to the none state
def __i2s_callback(self, arg):
# PLAY
if self.__state == WavPlayer.PLAY:
if self.__mode == WavPlayer.MODE_WAV:
num_read = self.__wav_file.readinto(self.__wav_samples_mv) # Read the next section of the WAV file
self.total_bytes_read += num_read
# Have we reached the end of the file?
if num_read == 0:
# Do we want to loop the WAV playback?
if self.__loop_wav:
_ = self.__wav_file.seek(self.__first_sample_offset) # Play again, so advance to first byte of sample data
else:
self.__wav_file.close() # Stop playing, so close the file
self.__state = WavPlayer.FLUSH # and enter the flush state on the next callback
self.__audio_out.write(self.__silence_samples) # In both cases play silence to end this callback
else:
if num_read > 0 and num_read < self.WAV_BUFFER_LENGTH:
num_read = num_read - (self.total_bytes_read - self.sample_size)
self.__audio_out.write(self.__wav_samples_mv[: num_read]) # We are within the file, so write out the next audio samples
else:
if self.__queued_samples is not None:
self.__tone_samples = self.__queued_samples
self.__queued_samples = None
self.__audio_out.write(self.__tone_samples)
# PAUSE or STOP
elif self.__state == WavPlayer.PAUSE or self.__state == WavPlayer.STOP:
self.__audio_out.write(self.__silence_samples) # Play silence
# FLUSH
elif self.__state == WavPlayer.FLUSH:
# Flush is used to allow the residual audio samples in the internal buffer to be written
# to the I2S peripheral. This step avoids part of the sound file from being cut off
if self.__flush_count > 0:
self.__flush_count -= 1
else:
self.__state = WavPlayer.STOP # Enter the stop state on the next callback
self.__audio_out.write(self.__silence_samples) # Play silence
# NONE
elif self.__state == WavPlayer.NONE:
pass
@staticmethod
def __parse_wav(wav_file):
chunk_ID = wav_file.read(4)
if chunk_ID != b"RIFF":
raise ValueError("WAV chunk ID invalid")
_ = wav_file.read(4) # chunk_size
format = wav_file.read(4)
if format != b"WAVE":
raise ValueError("WAV format invalid")
sub_chunk1_ID = wav_file.read(4)
if sub_chunk1_ID != b"fmt ":
raise ValueError("WAV sub chunk 1 ID invalid")
_ = wav_file.read(4) # sub_chunk1_size
_ = struct.unpack("<H", wav_file.read(2))[0] # audio_format
num_channels = struct.unpack("<H", wav_file.read(2))[0]
if num_channels == 1:
format = I2S.MONO
else:
format = I2S.STEREO
sample_rate = struct.unpack("<I", wav_file.read(4))[0]
# if sample_rate != 44_100 and sample_rate != 48_000:
# raise ValueError(f"WAV sample rate of {sample_rate} invalid. Only 44.1KHz or 48KHz audio are supported")
_ = struct.unpack("<I", wav_file.read(4))[0] # byte_rate
_ = struct.unpack("<H", wav_file.read(2))[0] # block_align
bits_per_sample = struct.unpack("<H", wav_file.read(2))[0]
# usually the sub chunk2 ID ("data") comes next, but
# some online MP3->WAV converters add
# binary data before "data". So, read a fairly large
# block of bytes and search for "data".
binary_block = wav_file.read(200)
offset = binary_block.find(b"data")
if offset == -1:
raise ValueError("WAV sub chunk 2 ID not found")
wav_file.seek(40)
sub_chunk2_size = struct.unpack("<I", wav_file.read(4))[0]
return (format, sample_rate, bits_per_sample, 44 + offset, sub_chunk2_size)

Plik binarny nie jest wyświetlany.

Plik binarny nie jest wyświetlany.

Wyświetl plik

@ -0,0 +1,86 @@
from machine import Timer
from audio import WavPlayer
from galactic import GalacticUnicorn
from picographics import PicoGraphics, DISPLAY_GALACTIC_UNICORN as DISPLAY
import time
gu = GalacticUnicorn()
graphics = PicoGraphics(DISPLAY)
graphics.set_font("bitmap6")
WHITE = graphics.create_pen(255, 255, 255)
BLUE = graphics.create_pen(0, 0, 255)
CLEAR = graphics.create_pen(0, 0, 0)
RED = graphics.create_pen(255, 0, 0)
GREEN = graphics.create_pen(0, 255, 0)
gu.set_brightness(0.7)
audio = WavPlayer(0, 10, 11, 9, amp_enable=22)
class Countdown(object):
def __init__(self):
self.timer_running = False
self.total_seconds = 0
self.timer = None
def process_input(self):
if gu.is_pressed(GalacticUnicorn.SWITCH_VOLUME_UP):
self.total_seconds += 1
if gu.is_pressed(GalacticUnicorn.SWITCH_VOLUME_DOWN):
if self.total_seconds > 0:
self.total_seconds -= 1
if gu.is_pressed(GalacticUnicorn.SWITCH_SLEEP):
self.start_timer()
def display_time(self):
seconds = self.total_seconds % (24 * 3600)
seconds %= 3600
minutes = seconds // 60
seconds %= 60
# Add leading zeros to the minutes and seconds
if len(str(minutes)) == 1:
minutes = "0{}".format(minutes)
if len(str(seconds)) == 1:
seconds = "0{}".format(seconds)
return "{}:{}".format(minutes, seconds)
def draw(self):
graphics.set_pen(graphics.create_pen(0, 0, 0))
graphics.clear()
graphics.set_pen(BLUE)
graphics.circle(0, 0, 12)
graphics.set_pen(GREEN)
graphics.circle(50, 7, 6)
graphics.set_pen(WHITE)
graphics.text(self.display_time(), 15, 2, -1, 1)
gu.update(graphics)
def start_timer(self):
if not self.timer_running:
self.timer = Timer(mode=Timer.PERIODIC, period=1000, callback=self.countdown)
self.timer_running = True
def reset(self):
self.timer.deinit()
self.timer_running = False
def countdown(self, arg):
if self.total_seconds == 0:
audio.play_wav("doorbell.wav", False)
self.reset()
else:
self.total_seconds -= 1
count = Countdown()
while 1:
count.process_input()
count.draw()
time.sleep(0.07)

Plik binarny nie jest wyświetlany.

Wyświetl plik

@ -0,0 +1,80 @@
from audio import WavPlayer
from galactic import GalacticUnicorn
from picographics import PicoGraphics, DISPLAY_GALACTIC_UNICORN as DISPLAY
from time import sleep
gu = GalacticUnicorn()
graphics = PicoGraphics(DISPLAY)
audio = WavPlayer(0, 10, 11, 9, amp_enable=22)
WHITE = graphics.create_pen(255, 255, 255)
RED = graphics.create_pen(255, 0, 0)
GREEN = graphics.create_pen(0, 255, 150)
CLEAR = graphics.create_pen(0, 0, 0)
class Menu(object):
def __init__(self):
self.items = ["Pew 1", "Pew 2", "Pew 3"]
self.selected = 0
# A function to draw only the menu elements.
# Helps to keep our main function tidy!
def draw_menu(self):
graphics.set_pen(WHITE)
graphics.set_pen(GREEN)
graphics.line(0, 5, GalacticUnicorn.WIDTH, 5)
graphics.set_pen(CLEAR)
graphics.rectangle(13, 2, 26, 5)
for item in range(len(self.items)):
if self.selected == item:
graphics.set_pen(WHITE)
graphics.text(self.items[self.selected], 14, 2, 31, 1)
# Make changes based on the currently selected menu item
def process_selected(self):
if self.selected == 0:
audio.play_wav("Pew1.wav", False)
if self.selected == 1:
audio.play_wav("Pew2.wav", False)
if self.selected == 2:
audio.play_wav("Pew3.wav", False)
menu = Menu()
graphics.set_font("bitmap6")
gu.set_brightness(0.7)
while True:
graphics.set_pen(graphics.create_pen(0, 0, 0))
graphics.clear()
if gu.is_pressed(GalacticUnicorn.SWITCH_BRIGHTNESS_DOWN):
audio.play_wav("buttonbeep.wav", False)
if menu.selected + 1 < len(menu.items):
menu.selected += 1
else:
menu.selected = 0
if gu.is_pressed(GalacticUnicorn.SWITCH_BRIGHTNESS_UP):
audio.play_wav("buttonbeep.wav", False)
if menu.selected > 0:
menu.selected -= 1
else:
menu.selected = len(menu.items) - 1
if gu.is_pressed(GalacticUnicorn.SWITCH_SLEEP):
menu.process_selected()
menu.draw_menu()
gu.update(graphics)
sleep(0.2)

Wyświetl plik

@ -0,0 +1,8 @@
from audio import WavPlayer
sound = WavPlayer(0, 10, 11, 9, amp_enable=22)
sound.play_wav("beepboop.wav", False)
while sound.is_playing():
pass

Wyświetl plik

@ -0,0 +1,276 @@
# SPDX-FileCopyrightText: 2023 Christopher Parrott for Pimoroni Ltd
#
# SPDX-License-Identifier: MIT
import os
import math
import struct
from machine import I2S, Pin
"""
A class for playing Wav files out of an I2S audio amp. It can also play pure tones.
This code is based heavily on the work of Mike Teachman, at:
https://github.com/miketeachman/micropython-i2s-examples/blob/master/examples/wavplayer.py
"""
class WavPlayer:
# Internal states
PLAY = 0
PAUSE = 1
FLUSH = 2
STOP = 3
NONE = 4
MODE_WAV = 0
MODE_TONE = 1
# Default buffer length
SILENCE_BUFFER_LENGTH = 1000
WAV_BUFFER_LENGTH = 10000
INTERNAL_BUFFER_LENGTH = 20000
TONE_SAMPLE_RATE = 44_100
TONE_BITS_PER_SAMPLE = 16
TONE_FULL_WAVES = 2
def __init__(self, id, sck_pin, ws_pin, sd_pin, amp_enable=None, ibuf_len=INTERNAL_BUFFER_LENGTH, root="/"):
self.__id = id
self.__sck_pin = sck_pin
self.__ws_pin = ws_pin
self.__sd_pin = sd_pin
self.__ibuf_len = ibuf_len
self.__enable = None
if amp_enable is not None:
self.__enable = Pin(amp_enable, Pin.OUT)
# Set the directory to search for files in
self.set_root(root)
self.__state = WavPlayer.NONE
self.__mode = WavPlayer.MODE_WAV
self.__wav_file = None
self.__loop_wav = False
self.__first_sample_offset = None
self.__flush_count = 0
self.__audio_out = None
# Allocate a small array of blank audio samples used for silence
self.__silence_samples = bytearray(self.SILENCE_BUFFER_LENGTH)
# Allocate a larger array for WAV audio samples, using a memoryview for more efficient access
self.__wav_samples_mv = memoryview(bytearray(self.WAV_BUFFER_LENGTH))
# Reserve a variable for audio samples used for tones
self.__tone_samples = None
self.__queued_samples = None
def set_root(self, root):
self.__root = root.rstrip("/") + "/"
def play_wav(self, wav_file, loop=False):
if os.listdir(self.__root).count(wav_file) == 0:
raise ValueError(f"'{wav_file}' not found")
self.__stop_i2s() # Stop any active playback and terminate the I2S instance
self.__wav_file = open(self.__root + wav_file, "rb") # Open the chosen WAV file in read-only, binary mode
self.__loop_wav = loop # Record if the user wants the file to loop
# Parse the WAV file, returning the necessary parameters to initialise I2S communication
format, sample_rate, bits_per_sample, self.__first_sample_offset, self.sample_size = WavPlayer.__parse_wav(self.__wav_file)
# Keep a track of total bytes read from WAV File
self.total_bytes_read = 0
self.__wav_file.seek(self.__first_sample_offset) # Advance to first byte of sample data
self.__start_i2s(bits=bits_per_sample,
format=format,
rate=sample_rate,
state=WavPlayer.PLAY,
mode=WavPlayer.MODE_WAV)
def play_tone(self, frequency, amplitude):
if frequency < 20.0 or frequency > 20_000:
raise ValueError("frequency out of range. Expected between 20Hz and 20KHz")
if amplitude < 0.0 or amplitude > 1.0:
raise ValueError("amplitude out of range. Expected 0.0 to 1.0")
# Create a buffer containing the pure tone samples
samples_per_cycle = self.TONE_SAMPLE_RATE // frequency
sample_size_in_bytes = self.TONE_BITS_PER_SAMPLE // 8
samples = bytearray(self.TONE_FULL_WAVES * samples_per_cycle * sample_size_in_bytes)
range = pow(2, self.TONE_BITS_PER_SAMPLE) // 2
format = "<h" if self.TONE_BITS_PER_SAMPLE == 16 else "<l"
# Populate the buffer with multiple cycles to avoid it completing too quickly and causing drop outs
for i in range(samples_per_cycle * self.TONE_FULL_WAVES):
sample = int((range - 1) * (math.sin(2 * math.pi * i / samples_per_cycle)) * amplitude)
struct.pack_into(format, samples, i * sample_size_in_bytes, sample)
# Are we not already playing tones?
if not (self.__mode == WavPlayer.MODE_TONE and (self.__state == WavPlayer.PLAY or self.__state == WavPlayer.PAUSE)):
self.__stop_i2s() # Stop any active playback and terminate the I2S instance
self.__tone_samples = samples
self.__start_i2s(bits=self.TONE_BITS_PER_SAMPLE,
format=I2S.MONO,
rate=self.TONE_SAMPLE_RATE,
state=WavPlayer.PLAY,
mode=WavPlayer.MODE_TONE)
else:
self.__queued_samples = samples
self.__state = WavPlayer.PLAY
def pause(self):
if self.__state == WavPlayer.PLAY:
self.__state = WavPlayer.PAUSE # Enter the pause state on the next callback
def resume(self):
if self.__state == WavPlayer.PAUSE:
self.__state = WavPlayer.PLAY # Enter the play state on the next callback
def stop(self):
if self.__state == WavPlayer.PLAY or self.__state == WavPlayer.PAUSE:
if self.__mode == WavPlayer.MODE_WAV:
# Enter the flush state on the next callback and close the file
# It is done in this order to prevent the callback entering the play
# state after we close the file but before we change the state)
self.__state = WavPlayer.FLUSH
self.__wav_file.close()
else:
self.__state = WavPlayer.STOP
def is_playing(self):
return self.__state != WavPlayer.NONE and self.__state != WavPlayer.STOP
def is_paused(self):
return self.__state == WavPlayer.PAUSE
def __start_i2s(self, bits=16, format=I2S.MONO, rate=44_100, state=STOP, mode=MODE_WAV):
import gc
gc.collect()
self.__audio_out = I2S(
self.__id,
sck=self.__sck_pin,
ws=self.__ws_pin,
sd=self.__sd_pin,
mode=I2S.TX,
bits=bits,
format=format,
rate=rate,
ibuf=self.__ibuf_len,
)
self.__state = state
self.__mode = mode
self.__flush_count = self.__ibuf_len // self.SILENCE_BUFFER_LENGTH + 1
self.__audio_out.irq(self.__i2s_callback)
self.__audio_out.write(self.__silence_samples)
if self.__enable is not None:
self.__enable.on()
def __stop_i2s(self):
self.stop() # Stop any active playback
while self.is_playing(): # and wait for it to complete
pass
if self.__enable is not None:
self.__enable.off()
if self.__audio_out is not None:
self.__audio_out.deinit() # Deinit any active I2S comms
self.__state == WavPlayer.NONE # Return to the none state
def __i2s_callback(self, arg):
# PLAY
if self.__state == WavPlayer.PLAY:
if self.__mode == WavPlayer.MODE_WAV:
num_read = self.__wav_file.readinto(self.__wav_samples_mv) # Read the next section of the WAV file
self.total_bytes_read += num_read
# Have we reached the end of the file?
if num_read == 0:
# Do we want to loop the WAV playback?
if self.__loop_wav:
_ = self.__wav_file.seek(self.__first_sample_offset) # Play again, so advance to first byte of sample data
else:
self.__wav_file.close() # Stop playing, so close the file
self.__state = WavPlayer.FLUSH # and enter the flush state on the next callback
self.__audio_out.write(self.__silence_samples) # In both cases play silence to end this callback
else:
if num_read > 0 and num_read < self.WAV_BUFFER_LENGTH:
num_read = num_read - (self.total_bytes_read - self.sample_size)
self.__audio_out.write(self.__wav_samples_mv[: num_read]) # We are within the file, so write out the next audio samples
else:
if self.__queued_samples is not None:
self.__tone_samples = self.__queued_samples
self.__queued_samples = None
self.__audio_out.write(self.__tone_samples)
# PAUSE or STOP
elif self.__state == WavPlayer.PAUSE or self.__state == WavPlayer.STOP:
self.__audio_out.write(self.__silence_samples) # Play silence
# FLUSH
elif self.__state == WavPlayer.FLUSH:
# Flush is used to allow the residual audio samples in the internal buffer to be written
# to the I2S peripheral. This step avoids part of the sound file from being cut off
if self.__flush_count > 0:
self.__flush_count -= 1
else:
self.__state = WavPlayer.STOP # Enter the stop state on the next callback
self.__audio_out.write(self.__silence_samples) # Play silence
# NONE
elif self.__state == WavPlayer.NONE:
pass
@staticmethod
def __parse_wav(wav_file):
chunk_ID = wav_file.read(4)
if chunk_ID != b"RIFF":
raise ValueError("WAV chunk ID invalid")
_ = wav_file.read(4) # chunk_size
format = wav_file.read(4)
if format != b"WAVE":
raise ValueError("WAV format invalid")
sub_chunk1_ID = wav_file.read(4)
if sub_chunk1_ID != b"fmt ":
raise ValueError("WAV sub chunk 1 ID invalid")
_ = wav_file.read(4) # sub_chunk1_size
_ = struct.unpack("<H", wav_file.read(2))[0] # audio_format
num_channels = struct.unpack("<H", wav_file.read(2))[0]
if num_channels == 1:
format = I2S.MONO
else:
format = I2S.STEREO
sample_rate = struct.unpack("<I", wav_file.read(4))[0]
# if sample_rate != 44_100 and sample_rate != 48_000:
# raise ValueError(f"WAV sample rate of {sample_rate} invalid. Only 44.1KHz or 48KHz audio are supported")
_ = struct.unpack("<I", wav_file.read(4))[0] # byte_rate
_ = struct.unpack("<H", wav_file.read(2))[0] # block_align
bits_per_sample = struct.unpack("<H", wav_file.read(2))[0]
# usually the sub chunk2 ID ("data") comes next, but
# some online MP3->WAV converters add
# binary data before "data". So, read a fairly large
# block of bytes and search for "data".
binary_block = wav_file.read(200)
offset = binary_block.find(b"data")
if offset == -1:
raise ValueError("WAV sub chunk 2 ID not found")
wav_file.seek(40)
sub_chunk2_size = struct.unpack("<I", wav_file.read(4))[0]
return (format, sample_rate, bits_per_sample, 44 + offset, sub_chunk2_size)

Plik binarny nie jest wyświetlany.

Wyświetl plik

@ -0,0 +1,8 @@
from audio import WavPlayer
sound = WavPlayer(0, 10, 11, 9, amp_enable=29)
sound.play_wav("pirate-arrrr.wav", False)
while sound.is_playing():
pass

Plik binarny nie jest wyświetlany.

Plik binarny nie jest wyświetlany.

Plik binarny nie jest wyświetlany.

Wyświetl plik

@ -0,0 +1,276 @@
# SPDX-FileCopyrightText: 2023 Christopher Parrott for Pimoroni Ltd
#
# SPDX-License-Identifier: MIT
import os
import math
import struct
from machine import I2S, Pin
"""
A class for playing Wav files out of an I2S audio amp. It can also play pure tones.
This code is based heavily on the work of Mike Teachman, at:
https://github.com/miketeachman/micropython-i2s-examples/blob/master/examples/wavplayer.py
"""
class WavPlayer:
# Internal states
PLAY = 0
PAUSE = 1
FLUSH = 2
STOP = 3
NONE = 4
MODE_WAV = 0
MODE_TONE = 1
# Default buffer length
SILENCE_BUFFER_LENGTH = 1000
WAV_BUFFER_LENGTH = 10000
INTERNAL_BUFFER_LENGTH = 20000
TONE_SAMPLE_RATE = 44_100
TONE_BITS_PER_SAMPLE = 16
TONE_FULL_WAVES = 2
def __init__(self, id, sck_pin, ws_pin, sd_pin, amp_enable=None, ibuf_len=INTERNAL_BUFFER_LENGTH, root="/"):
self.__id = id
self.__sck_pin = sck_pin
self.__ws_pin = ws_pin
self.__sd_pin = sd_pin
self.__ibuf_len = ibuf_len
self.__enable = None
if amp_enable is not None:
self.__enable = Pin(amp_enable, Pin.OUT)
# Set the directory to search for files in
self.set_root(root)
self.__state = WavPlayer.NONE
self.__mode = WavPlayer.MODE_WAV
self.__wav_file = None
self.__loop_wav = False
self.__first_sample_offset = None
self.__flush_count = 0
self.__audio_out = None
# Allocate a small array of blank audio samples used for silence
self.__silence_samples = bytearray(self.SILENCE_BUFFER_LENGTH)
# Allocate a larger array for WAV audio samples, using a memoryview for more efficient access
self.__wav_samples_mv = memoryview(bytearray(self.WAV_BUFFER_LENGTH))
# Reserve a variable for audio samples used for tones
self.__tone_samples = None
self.__queued_samples = None
def set_root(self, root):
self.__root = root.rstrip("/") + "/"
def play_wav(self, wav_file, loop=False):
if os.listdir(self.__root).count(wav_file) == 0:
raise ValueError(f"'{wav_file}' not found")
self.__stop_i2s() # Stop any active playback and terminate the I2S instance
self.__wav_file = open(self.__root + wav_file, "rb") # Open the chosen WAV file in read-only, binary mode
self.__loop_wav = loop # Record if the user wants the file to loop
# Parse the WAV file, returning the necessary parameters to initialise I2S communication
format, sample_rate, bits_per_sample, self.__first_sample_offset, self.sample_size = WavPlayer.__parse_wav(self.__wav_file)
# Keep a track of total bytes read from WAV File
self.total_bytes_read = 0
self.__wav_file.seek(self.__first_sample_offset) # Advance to first byte of sample data
self.__start_i2s(bits=bits_per_sample,
format=format,
rate=sample_rate,
state=WavPlayer.PLAY,
mode=WavPlayer.MODE_WAV)
def play_tone(self, frequency, amplitude):
if frequency < 20.0 or frequency > 20_000:
raise ValueError("frequency out of range. Expected between 20Hz and 20KHz")
if amplitude < 0.0 or amplitude > 1.0:
raise ValueError("amplitude out of range. Expected 0.0 to 1.0")
# Create a buffer containing the pure tone samples
samples_per_cycle = self.TONE_SAMPLE_RATE // frequency
sample_size_in_bytes = self.TONE_BITS_PER_SAMPLE // 8
samples = bytearray(self.TONE_FULL_WAVES * samples_per_cycle * sample_size_in_bytes)
range = pow(2, self.TONE_BITS_PER_SAMPLE) // 2
format = "<h" if self.TONE_BITS_PER_SAMPLE == 16 else "<l"
# Populate the buffer with multiple cycles to avoid it completing too quickly and causing drop outs
for i in range(samples_per_cycle * self.TONE_FULL_WAVES):
sample = int((range - 1) * (math.sin(2 * math.pi * i / samples_per_cycle)) * amplitude)
struct.pack_into(format, samples, i * sample_size_in_bytes, sample)
# Are we not already playing tones?
if not (self.__mode == WavPlayer.MODE_TONE and (self.__state == WavPlayer.PLAY or self.__state == WavPlayer.PAUSE)):
self.__stop_i2s() # Stop any active playback and terminate the I2S instance
self.__tone_samples = samples
self.__start_i2s(bits=self.TONE_BITS_PER_SAMPLE,
format=I2S.MONO,
rate=self.TONE_SAMPLE_RATE,
state=WavPlayer.PLAY,
mode=WavPlayer.MODE_TONE)
else:
self.__queued_samples = samples
self.__state = WavPlayer.PLAY
def pause(self):
if self.__state == WavPlayer.PLAY:
self.__state = WavPlayer.PAUSE # Enter the pause state on the next callback
def resume(self):
if self.__state == WavPlayer.PAUSE:
self.__state = WavPlayer.PLAY # Enter the play state on the next callback
def stop(self):
if self.__state == WavPlayer.PLAY or self.__state == WavPlayer.PAUSE:
if self.__mode == WavPlayer.MODE_WAV:
# Enter the flush state on the next callback and close the file
# It is done in this order to prevent the callback entering the play
# state after we close the file but before we change the state)
self.__state = WavPlayer.FLUSH
self.__wav_file.close()
else:
self.__state = WavPlayer.STOP
def is_playing(self):
return self.__state != WavPlayer.NONE and self.__state != WavPlayer.STOP
def is_paused(self):
return self.__state == WavPlayer.PAUSE
def __start_i2s(self, bits=16, format=I2S.MONO, rate=44_100, state=STOP, mode=MODE_WAV):
import gc
gc.collect()
self.__audio_out = I2S(
self.__id,
sck=self.__sck_pin,
ws=self.__ws_pin,
sd=self.__sd_pin,
mode=I2S.TX,
bits=bits,
format=format,
rate=rate,
ibuf=self.__ibuf_len,
)
self.__state = state
self.__mode = mode
self.__flush_count = self.__ibuf_len // self.SILENCE_BUFFER_LENGTH + 1
self.__audio_out.irq(self.__i2s_callback)
self.__audio_out.write(self.__silence_samples)
if self.__enable is not None:
self.__enable.on()
def __stop_i2s(self):
self.stop() # Stop any active playback
while self.is_playing(): # and wait for it to complete
pass
if self.__enable is not None:
self.__enable.off()
if self.__audio_out is not None:
self.__audio_out.deinit() # Deinit any active I2S comms
self.__state == WavPlayer.NONE # Return to the none state
def __i2s_callback(self, arg):
# PLAY
if self.__state == WavPlayer.PLAY:
if self.__mode == WavPlayer.MODE_WAV:
num_read = self.__wav_file.readinto(self.__wav_samples_mv) # Read the next section of the WAV file
self.total_bytes_read += num_read
# Have we reached the end of the file?
if num_read == 0:
# Do we want to loop the WAV playback?
if self.__loop_wav:
_ = self.__wav_file.seek(self.__first_sample_offset) # Play again, so advance to first byte of sample data
else:
self.__wav_file.close() # Stop playing, so close the file
self.__state = WavPlayer.FLUSH # and enter the flush state on the next callback
self.__audio_out.write(self.__silence_samples) # In both cases play silence to end this callback
else:
if num_read > 0 and num_read < self.WAV_BUFFER_LENGTH:
num_read = num_read - (self.total_bytes_read - self.sample_size)
self.__audio_out.write(self.__wav_samples_mv[: num_read]) # We are within the file, so write out the next audio samples
else:
if self.__queued_samples is not None:
self.__tone_samples = self.__queued_samples
self.__queued_samples = None
self.__audio_out.write(self.__tone_samples)
# PAUSE or STOP
elif self.__state == WavPlayer.PAUSE or self.__state == WavPlayer.STOP:
self.__audio_out.write(self.__silence_samples) # Play silence
# FLUSH
elif self.__state == WavPlayer.FLUSH:
# Flush is used to allow the residual audio samples in the internal buffer to be written
# to the I2S peripheral. This step avoids part of the sound file from being cut off
if self.__flush_count > 0:
self.__flush_count -= 1
else:
self.__state = WavPlayer.STOP # Enter the stop state on the next callback
self.__audio_out.write(self.__silence_samples) # Play silence
# NONE
elif self.__state == WavPlayer.NONE:
pass
@staticmethod
def __parse_wav(wav_file):
chunk_ID = wav_file.read(4)
if chunk_ID != b"RIFF":
raise ValueError("WAV chunk ID invalid")
_ = wav_file.read(4) # chunk_size
format = wav_file.read(4)
if format != b"WAVE":
raise ValueError("WAV format invalid")
sub_chunk1_ID = wav_file.read(4)
if sub_chunk1_ID != b"fmt ":
raise ValueError("WAV sub chunk 1 ID invalid")
_ = wav_file.read(4) # sub_chunk1_size
_ = struct.unpack("<H", wav_file.read(2))[0] # audio_format
num_channels = struct.unpack("<H", wav_file.read(2))[0]
if num_channels == 1:
format = I2S.MONO
else:
format = I2S.STEREO
sample_rate = struct.unpack("<I", wav_file.read(4))[0]
# if sample_rate != 44_100 and sample_rate != 48_000:
# raise ValueError(f"WAV sample rate of {sample_rate} invalid. Only 44.1KHz or 48KHz audio are supported")
_ = struct.unpack("<I", wav_file.read(4))[0] # byte_rate
_ = struct.unpack("<H", wav_file.read(2))[0] # block_align
bits_per_sample = struct.unpack("<H", wav_file.read(2))[0]
# usually the sub chunk2 ID ("data") comes next, but
# some online MP3->WAV converters add
# binary data before "data". So, read a fairly large
# block of bytes and search for "data".
binary_block = wav_file.read(200)
offset = binary_block.find(b"data")
if offset == -1:
raise ValueError("WAV sub chunk 2 ID not found")
wav_file.seek(40)
sub_chunk2_size = struct.unpack("<I", wav_file.read(4))[0]
return (format, sample_rate, bits_per_sample, 44 + offset, sub_chunk2_size)

Plik binarny nie jest wyświetlany.

Plik binarny nie jest wyświetlany.

Wyświetl plik

@ -0,0 +1,93 @@
# Countdown Timer with Alarm
# i2s Audio Example
# Use VOL +/- to increase/decrease the amount of time
# Use the Sleep ZZZ button to start the countdown
from machine import Timer
from audio import WavPlayer
from stellar import StellarUnicorn
from picographics import PicoGraphics, DISPLAY_STELLAR_UNICORN as DISPLAY
import time
su = StellarUnicorn()
graphics = PicoGraphics(DISPLAY)
graphics.set_font("bitmap6")
WHITE = graphics.create_pen(255, 255, 255)
BLUE = graphics.create_pen(0, 0, 255)
CLEAR = graphics.create_pen(0, 0, 0)
RED = graphics.create_pen(255, 0, 0)
GREEN = graphics.create_pen(0, 255, 0)
su.set_brightness(0.5)
audio = WavPlayer(0, 10, 11, 9, amp_enable=22)
class Countdown(object):
def __init__(self):
self.timer_running = False
self.total_seconds = 0
self.timer = None
def process_input(self):
if su.is_pressed(StellarUnicorn.SWITCH_VOLUME_UP):
self.total_seconds += 1
if su.is_pressed(StellarUnicorn.SWITCH_VOLUME_DOWN):
if self.total_seconds > 0:
self.total_seconds -= 1
if su.is_pressed(StellarUnicorn.SWITCH_SLEEP):
self.start_timer()
def display_time(self):
seconds = self.total_seconds % (24 * 3600)
seconds %= 3600
minutes = seconds // 60
seconds %= 60
# Add leading zeros to the minutes and seconds
if len(str(minutes)) == 1:
minutes = "0{}".format(minutes)
if len(str(seconds)) == 1:
seconds = "0{}".format(seconds)
return "{}\n{}".format(minutes, seconds)
def draw(self):
graphics.set_pen(graphics.create_pen(0, 0, 0))
graphics.clear()
graphics.set_pen(RED)
graphics.rectangle(0, 0, 16, 16)
graphics.set_pen(CLEAR)
graphics.rectangle(2, 2, StellarUnicorn.WIDTH - 4, 12)
graphics.set_pen(WHITE)
graphics.text(self.display_time(), 2, 1, -1, 1)
su.update(graphics)
def start_timer(self):
if not self.timer_running:
self.timer = Timer(mode=Timer.PERIODIC, period=1000, callback=self.countdown)
self.timer_running = True
def reset(self):
time.sleep(0.2)
self.timer.deinit()
self.timer_running = False
def countdown(self, arg):
if self.total_seconds == 0:
audio.play_wav("doorbell.wav", False)
self.reset()
else:
self.total_seconds -= 1
count = Countdown()
while 1:
count.process_input()
count.draw()
time.sleep(0.07)

Plik binarny nie jest wyświetlany.

Wyświetl plik

@ -0,0 +1,86 @@
# Example Menu with Sound
# i2s Audio Example
# Use Brightness +/- to move up and down
# Press Sleep to play the selected sound
from audio import WavPlayer
from stellar import StellarUnicorn
from picographics import PicoGraphics, DISPLAY_STELLAR_UNICORN as DISPLAY
from time import sleep
su = StellarUnicorn()
graphics = PicoGraphics(DISPLAY)
audio = WavPlayer(0, 10, 11, 9, amp_enable=22)
WHITE = graphics.create_pen(255, 255, 255)
RED = graphics.create_pen(255, 0, 0)
GREEN = graphics.create_pen(0, 200, 200)
class Menu(object):
def __init__(self):
self.items = ["1", "2", "3"]
self.selected = 0
# A function to draw only the menu elements.
# Helps to keep our main function tidy!
def draw_menu(self):
graphics.set_pen(GREEN)
graphics.circle(0, 0, 3)
graphics.circle(16, 5, 4)
graphics.circle(3, 16, 3)
graphics.set_pen(WHITE)
for item in range(len(self.items)):
if self.selected == item:
graphics.set_pen(WHITE)
graphics.text(self.items[self.selected], 6, 4, 31, 1)
# Make changes based on the currently selected menu item
def process_selected(self):
if self.selected == 0:
audio.play_wav("Pew1.wav", False)
if self.selected == 1:
audio.play_wav("Pew2.wav", False)
if self.selected == 2:
audio.play_wav("Pew3.wav", False)
menu = Menu()
graphics.set_font("bitmap6")
su.set_brightness(0.7)
while True:
graphics.set_pen(graphics.create_pen(0, 0, 0))
graphics.clear()
if su.is_pressed(StellarUnicorn.SWITCH_BRIGHTNESS_DOWN):
audio.play_wav("buttonbeep.wav", False)
if menu.selected + 1 < len(menu.items):
menu.selected += 1
else:
menu.selected = 0
if su.is_pressed(StellarUnicorn.SWITCH_BRIGHTNESS_UP):
audio.play_wav("buttonbeep.wav", False)
if menu.selected > 0:
menu.selected -= 1
else:
menu.selected = len(menu.items) - 1
if su.is_pressed(StellarUnicorn.SWITCH_SLEEP):
menu.process_selected()
menu.draw_menu()
su.update(graphics)
sleep(0.2)

Wyświetl plik

@ -0,0 +1,8 @@
from audio import WavPlayer
sound = WavPlayer(0, 10, 11, 9, amp_enable=22)
sound.play_wav("beepboop.wav", False)
while sound.is_playing():
pass