音频流Python上的Google流语音识别

Question

我已经搜索了Google的所有可用文档，但我找不到Python音频流上的流式语音识别示例。

目前，我正在Django中使用Python语音识别从用户那里获取音频，然后收听音频。然后，我可以保存文件并运行google语音识别，或者直接从创建的音频实例中运行。

有人能指导我如何对音频流执行流式语音识别吗？

Answer 1

如果您使用React web app来流式传输客户端的音频，那么您可以参考此存储库以获取代码示例（或者您可以克隆它并添加您的专有代码）https://github.com/saharmor/realtime-transcription-playground

Answer 2

这是上述要求的工作代码。

代码：

import asyncio
import websockets
import json
import threading
from six.moves import queue
from google.cloud import speech
from google.cloud.speech import types


IP = '0.0.0.0'
PORT = 8000

class Transcoder(object):
    """
    Converts audio chunks to text
    """
    def __init__(self, encoding, rate, language):
        self.buff = queue.Queue()
        self.encoding = encoding
        self.language = language
        self.rate = rate
        self.closed = True
        self.transcript = None

    def start(self):
        """Start up streaming speech call"""
        threading.Thread(target=self.process).start()

    def response_loop(self, responses):
        """
        Pick up the final result of Speech to text conversion
        """
        for response in responses:
            if not response.results:
                continue
            result = response.results[0]
            if not result.alternatives:
                continue
            transcript = result.alternatives[0].transcript
            if result.is_final:
                self.transcript = transcript

    def process(self):
        """
        Audio stream recognition and result parsing
        """
        #You can add speech contexts for better recognition
        cap_speech_context = types.SpeechContext(phrases=["Add your phrases here"])
        client = speech.SpeechClient()
        config = types.RecognitionConfig(
            encoding=self.encoding,
            sample_rate_hertz=self.rate,
            language_code=self.language,
            speech_contexts=[cap_speech_context,],
            model='command_and_search'
        )
        streaming_config = types.StreamingRecognitionConfig(
            config=config,
            interim_results=False,
            single_utterance=False)
        audio_generator = self.stream_generator()
        requests = (types.StreamingRecognizeRequest(audio_content=content)
                    for content in audio_generator)

        responses = client.streaming_recognize(streaming_config, requests)
        try:
            self.response_loop(responses)
        except:
            self.start()

    def stream_generator(self):
        while not self.closed:
            chunk = self.buff.get()
            if chunk is None:
                return
            data = [chunk]
            while True:
                try:
                    chunk = self.buff.get(block=False)
                    if chunk is None:
                        return
                    data.append(chunk)
                except queue.Empty:
                    break
            yield b''.join(data)

    def write(self, data):
        """
        Writes data to the buffer
        """
        self.buff.put(data)


async def audio_processor(websocket, path):
    """
    Collects audio from the stream, writes it to buffer and return the output of Google speech to text
    """
    config = await websocket.recv()
    if not isinstance(config, str):
        print("ERROR, no config")
        return
    config = json.loads(config)
    transcoder = Transcoder(
        encoding=config["format"],
        rate=config["rate"],
        language=config["language"]
    )
    transcoder.start()
    while True:
        try:
            data = await websocket.recv()
        except websockets.ConnectionClosed:
            print("Connection closed")
            break
        transcoder.write(data)
        transcoder.closed = False
        if transcoder.transcript:
            print(transcoder.transcript)
            await websocket.send(transcoder.transcript)
            transcoder.transcript = None

start_server = websockets.serve(audio_processor, IP, PORT)
asyncio.get_event_loop().run_until_complete(start_server)
asyncio.get_event_loop().run_forever()

Answer 3

Google在这里提供了一个流式Python API的示例。

不要打开音频文件来创建流（如该示例的第34行所示），而是将流直接传递给音频样本对象（如第36行所示）。