mirror of
https://github.com/mozilla/DeepSpeech.git
synced 2025-10-26 11:19:39 +00:00
188 lines
8.2 KiB
Python
188 lines
8.2 KiB
Python
import time, logging
|
|
from datetime import datetime
|
|
import threading, collections, queue, os, os.path
|
|
import wave
|
|
import pyaudio
|
|
import webrtcvad
|
|
from halo import Halo
|
|
import deepspeech
|
|
import numpy as np
|
|
|
|
logging.basicConfig(level=20)
|
|
|
|
class Audio(object):
|
|
"""Streams raw audio from microphone. Data is received in a separate thread, and stored in a buffer, to be read from."""
|
|
|
|
FORMAT = pyaudio.paInt16
|
|
RATE = 16000
|
|
CHANNELS = 1
|
|
BLOCKS_PER_SECOND = 50
|
|
BLOCK_SIZE = int(RATE / float(BLOCKS_PER_SECOND))
|
|
|
|
def __init__(self, callback=None):
|
|
def proxy_callback(in_data, frame_count, time_info, status):
|
|
callback(in_data)
|
|
return (None, pyaudio.paContinue)
|
|
if callback is None: callback = lambda in_data: self.buffer_queue.put(in_data)
|
|
self.buffer_queue = queue.Queue()
|
|
self.sample_rate = self.RATE
|
|
self.block_size = self.BLOCK_SIZE
|
|
self.pa = pyaudio.PyAudio()
|
|
self.stream = self.pa.open(format=self.FORMAT,
|
|
channels=self.CHANNELS,
|
|
rate=self.sample_rate,
|
|
input=True,
|
|
frames_per_buffer=self.block_size,
|
|
stream_callback=proxy_callback)
|
|
self.stream.start_stream()
|
|
|
|
def read(self):
|
|
"""Return a block of audio data, blocking if necessary."""
|
|
return self.buffer_queue.get()
|
|
|
|
def destroy(self):
|
|
self.stream.stop_stream()
|
|
self.stream.close()
|
|
self.pa.terminate()
|
|
|
|
frame_duration_ms = property(lambda self: 1000 * self.block_size // self.sample_rate)
|
|
|
|
def write_wav(self, filename, data):
|
|
logging.info("write wav %s", filename)
|
|
wf = wave.open(filename, 'wb')
|
|
wf.setnchannels(self.CHANNELS)
|
|
# wf.setsampwidth(self.pa.get_sample_size(FORMAT))
|
|
assert self.FORMAT == pyaudio.paInt16
|
|
wf.setsampwidth(2)
|
|
wf.setframerate(self.sample_rate)
|
|
wf.writeframes(data)
|
|
wf.close()
|
|
|
|
class VADAudio(Audio):
|
|
"""Filter & segment audio with voice activity detection."""
|
|
|
|
def __init__(self, aggressiveness=3):
|
|
super().__init__()
|
|
self.vad = webrtcvad.Vad(aggressiveness)
|
|
|
|
def frame_generator(self):
|
|
"""Generator that yields all audio frames from microphone."""
|
|
while True:
|
|
yield self.read()
|
|
|
|
def vad_collector(self, padding_ms=300, ratio=0.75, frames=None):
|
|
"""Generator that yields series of consecutive audio frames comprising each utterence, separated by yielding a single None.
|
|
Determines voice activity by ratio of frames in padding_ms. Uses a buffer to include padding_ms prior to being triggered.
|
|
Example: (frame, ..., frame, None, frame, ..., frame, None, ...)
|
|
|---utterence---| |---utterence---|
|
|
"""
|
|
if frames is None: frames = self.frame_generator()
|
|
num_padding_frames = padding_ms // self.frame_duration_ms
|
|
ring_buffer = collections.deque(maxlen=num_padding_frames)
|
|
triggered = False
|
|
|
|
for frame in frames:
|
|
is_speech = self.vad.is_speech(frame, self.sample_rate)
|
|
|
|
if not triggered:
|
|
ring_buffer.append((frame, is_speech))
|
|
num_voiced = len([f for f, speech in ring_buffer if speech])
|
|
if num_voiced > ratio * ring_buffer.maxlen:
|
|
triggered = True
|
|
for f, s in ring_buffer:
|
|
yield f
|
|
ring_buffer.clear()
|
|
|
|
else:
|
|
yield frame
|
|
ring_buffer.append((frame, is_speech))
|
|
num_unvoiced = len([f for f, speech in ring_buffer if not speech])
|
|
if num_unvoiced > ratio * ring_buffer.maxlen:
|
|
triggered = False
|
|
yield None
|
|
ring_buffer.clear()
|
|
|
|
def main(ARGS):
|
|
# Load DeepSpeech model
|
|
if os.path.isdir(ARGS.model):
|
|
model_dir = ARGS.model
|
|
ARGS.model = os.path.join(model_dir, 'output_graph.pb')
|
|
ARGS.alphabet = os.path.join(model_dir, ARGS.alphabet if ARGS.alphabet else 'alphabet.txt')
|
|
ARGS.lm = os.path.join(model_dir, ARGS.lm)
|
|
ARGS.trie = os.path.join(model_dir, ARGS.trie)
|
|
|
|
print('Initializing model...')
|
|
logging.info("ARGS.model: %s", ARGS.model)
|
|
logging.info("ARGS.alphabet: %s", ARGS.alphabet)
|
|
model = deepspeech.Model(ARGS.model, ARGS.n_features, ARGS.n_context, ARGS.alphabet, ARGS.beam_width)
|
|
if ARGS.lm and ARGS.trie:
|
|
logging.info("ARGS.lm: %s", ARGS.lm)
|
|
logging.info("ARGS.trie: %s", ARGS.trie)
|
|
model.enableDecoderWithLM(ARGS.alphabet, ARGS.lm, ARGS.trie, ARGS.lm_alpha, ARGS.lm_beta)
|
|
|
|
# Start audio with VAD
|
|
vad_audio = VADAudio(aggressiveness=ARGS.vad_aggressiveness)
|
|
print("Listening (ctrl-C to exit)...")
|
|
frames = vad_audio.vad_collector()
|
|
|
|
# Stream from microphone to DeepSpeech using VAD
|
|
spinner = None
|
|
if not ARGS.nospinner: spinner = Halo(spinner='line')
|
|
stream_context = model.setupStream()
|
|
wav_data = bytearray()
|
|
for frame in frames:
|
|
if frame is not None:
|
|
if spinner: spinner.start()
|
|
logging.debug("streaming frame")
|
|
model.feedAudioContent(stream_context, np.frombuffer(frame, np.int16))
|
|
if ARGS.savewav: wav_data.extend(frame)
|
|
else:
|
|
if spinner: spinner.stop()
|
|
logging.debug("end utterence")
|
|
if ARGS.savewav:
|
|
vad_audio.write_wav(os.path.join(ARGS.savewav, datetime.now().strftime("savewav_%Y-%m-%d_%H-%M-%S_%f.wav")), wav_data)
|
|
wav_data = bytearray()
|
|
text = model.finishStream(stream_context)
|
|
print("Recognized: %s" % text)
|
|
stream_context = model.setupStream()
|
|
|
|
if __name__ == '__main__':
|
|
BEAM_WIDTH = 500
|
|
LM_ALPHA = 0.75
|
|
LM_BETA = 1.85
|
|
N_FEATURES = 26
|
|
N_CONTEXT = 9
|
|
|
|
import argparse
|
|
parser = argparse.ArgumentParser(description="Stream from microphone to DeepSpeech using VAD")
|
|
|
|
parser.add_argument('-v', '--vad_aggressiveness', type=int, default=3,
|
|
help="Set aggressiveness of VAD: an integer between 0 and 3, 0 being the least aggressive about filtering out non-speech, 3 the most aggressive. Default: 3")
|
|
parser.add_argument('--nospinner', action='store_true',
|
|
help="Disable spinner")
|
|
parser.add_argument('-w', '--savewav',
|
|
help="Save .wav files of utterences to given directory")
|
|
|
|
parser.add_argument('-m', '--model', required=True,
|
|
help="Path to the model (protocol buffer binary file, or entire directory containing all standard-named files for model)")
|
|
parser.add_argument('-a', '--alphabet', default='alphabet.txt',
|
|
help="Path to the configuration file specifying the alphabet used by the network. Default: alphabet.txt")
|
|
parser.add_argument('-l', '--lm', default='lm.binary',
|
|
help="Path to the language model binary file. Default: lm.binary")
|
|
parser.add_argument('-t', '--trie', default='trie',
|
|
help="Path to the language model trie file created with native_client/generate_trie. Default: trie")
|
|
parser.add_argument('-nf', '--n_features', type=int, default=N_FEATURES,
|
|
help=f"Number of MFCC features to use. Default: {N_FEATURES}")
|
|
parser.add_argument('-nc', '--n_context', type=int, default=N_CONTEXT,
|
|
help=f"Size of the context window used for producing timesteps in the input vector. Default: {N_CONTEXT}")
|
|
parser.add_argument('-la', '--lm_alpha', type=float, default=LM_ALPHA,
|
|
help=f"The alpha hyperparameter of the CTC decoder. Language Model weight. Default: {LM_ALPHA}")
|
|
parser.add_argument('-lb', '--lm_beta', type=float, default=LM_BETA,
|
|
help=f"The beta hyperparameter of the CTC decoder. Word insertion bonus. Default: {LM_BETA}")
|
|
parser.add_argument('-bw', '--beam_width', type=int, default=BEAM_WIDTH,
|
|
help=f"Beam width used in the CTC decoder when building candidate transcriptions. Default: {BEAM_WIDTH}")
|
|
|
|
ARGS = parser.parse_args()
|
|
if ARGS.savewav: os.makedirs(ARGS.savewav, exist_ok=True)
|
|
main(ARGS)
|