时间戳

首页 > 解决方案 > 在 AWS Transcribe 中实时获取 BadRequestException

python - 在 AWS Transcribe 中实时获取 BadRequestException

问题描述

我从 amazon streaming api 得到这个响应。任何人都可以帮我解决我在这里做错了什么。

b'\x00\x00\x00\xa3\x00\x00\x00ah\x10k\xe1\x0f:exception-type\x07\x00\x13BadRequestException\r:content-type\x07\x00\x10application/json\r:message -type\x07\x00\textception{"Message":"收到意外的 WebSocket 帧。"}\xbd\xceK\x8a'

:message-type exception{"Message":"收到意外的 WebSocket 帧。"}½ÎK

我正在使用下面的代码

导入所有库

import asyncio
import websockets
import json
import sys, os, base64, datetime, hashlib, hmac, urllib
import pyaudio
import struct
import numpy as np
import wave
import argparse
import tempfile
import queue
import sys
import sounddevice as sd
import soundfile as sf
import numpy  # Make sure NumPy is loaded before it is used in the callback
assert numpy  # avoid "imported but unused" message (W0611)

使用 Sounddevice 流式传输音频的代码

def int_or_str(text):
    """Helper function for argument parsing."""
    try:
        return int(text)
    except ValueError:
        return text


parser = argparse.ArgumentParser(add_help=False)
parser.add_argument('-l', '--list-devices', action='store_true', help='show list of audio devices and exit')
args, remaining = parser.parse_known_args()
if args.list_devices:
    print(sd.query_devices())
    parser.exit(0)
parser = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter, parents=[parser])
parser.add_argument('filename', nargs='?', metavar='FILENAME',help='audio file to store recording to')
parser.add_argument('-d', '--device', type=int_or_str,help='input device (numeric ID or substring)')
parser.add_argument('-r', '--samplerate', type=int, help='sampling rate')
parser.add_argument('-c', '--channels', type=int, default=1, help='number of input channels')
parser.add_argument('-t', '--subtype', type=str, help='sound file subtype (e.g. "PCM_24")')
args = parser.parse_args(remaining)

q = queue.Queue()



def callback(indata, frames, time, status):
    """This is called (from a separate thread) for each audio block."""
    if status:
        print(status, file=sys.stderr)
    q.put(indata.copy())

根据亚马逊转录的要求创建用于连接的 URl

def createPresignedUrl(data):
    method = 'GET'
    service = 'transcribe'
    region = data['region']
    host = 'transcribestreaming.' + region + '.amazonaws.com:8443'
    endpoint = "wss://" + host

    def sign(key, msg):
        return hmac.new(key, msg.encode('utf-8'), hashlib.sha256).digest()

    def getSignatureKey(key, dateStamp, regionName, serviceName):
        kDate = sign(('AWS4' + key).encode('utf-8'), dateStamp)
        kRegion = sign(kDate, regionName)
        kService = sign(kRegion, serviceName)
        kSigning = sign(kService, 'aws4_request')
        return kSigning

    access_key = data['key']
    secret_key = data['secret']
    if access_key == '' or secret_key == '':
        print('No access key is available.')
        sys.exit()

    # Create a date for headers and the credential string
    t = datetime.datetime.utcnow()
    amz_date = t.strftime('%Y%m%dT%H%M%SZ') # Format date as YYYYMMDD'T'HHMMSS'Z'
    datestamp = t.strftime('%Y%m%d') # Date w/o time, used in credential scope

    canonical_uri = '/stream-transcription-websocket'

    canonical_headers = 'host:' + host + '\n'
    signed_headers = 'host'

    algorithm = 'AWS4-HMAC-SHA256'
    credential_scope = datestamp + '/' + region + '/' + service + '/' + 'aws4_request'

    canonical_querystring  = 'X-Amz-Algorithm=AWS4-HMAC-SHA256'
    canonical_querystring += '&X-Amz-Credential=' + urllib.parse.quote_plus(access_key + '/' + credential_scope)
    canonical_querystring += '&X-Amz-Date=' + amz_date
    canonical_querystring += '&X-Amz-Expires=300'
    canonical_querystring += '&X-Amz-SignedHeaders=' + signed_headers
    canonical_querystring += '&language-code=' + data['languageCode']
    canonical_querystring += '&media-encoding=pcm'
    canonical_querystring += '&sample-rate=' + str(data['sampleRate'])

    payload_hash = hashlib.sha256(('').encode('utf-8')).hexdigest()

    # Step 6: Combine elements to create canonical request
    canonical_request = method + '\n' + canonical_uri + '\n' + canonical_querystring + '\n' + canonical_headers + '\n' + signed_headers + '\n' + payload_hash

    string_to_sign = algorithm + '\n' +  amz_date + '\n' +  credential_scope + '\n' +  hashlib.sha256(canonical_request.encode('utf-8')).hexdigest()

    # Create the signing key
    signing_key = getSignatureKey(secret_key, datestamp, region, service)

    # Sign the string_to_sign using the signing_key
    signature = hmac.new(signing_key, (string_to_sign).encode("utf-8"), hashlib.sha256).hexdigest()

    canonical_querystring += '&X-Amz-Signature=' + signature

    request_url = endpoint + canonical_uri + "?" + canonical_querystring

    return request_url

data = {
    'key': 'Add your key',
    'secret': 'Add your secret key',
    'region': 'us-east-1',
    'languageCode': 'en-US',
    'sampleRate': 44100
}

进行 PCM 编码的代码

url = createPresignedUrl(data)

# FORMAT = pyaudio.paInt16
CHANNELS = 1
RATE = 44100
CHUNK = 16000
frames = []

# stream = audio.open(format=FORMAT, channels=CHANNELS, rate=RATE, input=True, frames_per_buffer=CHUNK)

def pcmEncode(in_data):

    offset = 0
    input_len = len(in_data)
    buffer = []

    for i in range(input_len):
        offset += 2
        s = max(-1, min(1, in_data[i]))
        b = (s * 32768) if (s < 0) else (s * 32767)
        buffer.insert(offset, b)
    buffer = np.array(buffer)
    return buffer

def downsampleBuffer(buffer, outputSampleRate = 16000) :
    if outputSampleRate == RATE :
        return buffer
    sampleRateRatio = RATE / outputSampleRate
    newLength = round( len(buffer) / sampleRateRatio )
    result = []
    offsetResult = 0
    offsetBuffer = 0

    while offsetResult < newLength :
        nextOffsetBuffer = round((offsetResult + 1) * sampleRateRatio)
        accum = 0
        count = 0

        len_buffer = nextOffsetBuffer if nextOffsetBuffer < len(buffer) else len(buffer)

        for i in range( offsetBuffer, len_buffer):
            accum += buffer[i]
            count += 1

        result.append(accum / count)
        offsetResult += 1
        offsetBuffer = nextOffsetBuffer


    return result

与 AWS 建立连接以进行转录

async def start_stream():

    try:
        connection = websockets.connect(url)
        stream = sd.Stream(samplerate=args.samplerate, blocksize=None, device=args.device,
                           channels=args.channels, dtype=None, latency=None, extra_settings=None,
                           callback=None, finished_callback=None, clip_off=None, dither_off=None,
                           never_drop_input=None, prime_output_buffers_using_stream_callback=None)
        stream.start()
        while True:

            a = stream.read(200)
            buffer = downsampleBuffer(a[0])
            result = pcmEncode(buffer)
            async with connection as ws:
                sent_data = {
                    "headers": {
                        ":message-type": {"type": "string", "value": "event"},
                        ":event-type": {"type": "string", "value": "AudioEvent"},
                        ":content-type": {"type": "string", "value": "application/octet-stream"}
                    },
                    "body": str(result)
                }
                await ws.send(json.dumps(sent_data))
                response = await ws.recv()
                print(response)
                print(response.decode('latin1'))

    except KeyboardInterrupt:

        parser.exit('\nInterrupted by user')

    except Exception as e:

        parser.exit(type(e).__name__ + ': ' + str(e))

    if status:
        parser.exit('Error during playback: ' + str(status))

asyncio.get_event_loop().run_until_complete(start_stream())
stream.stop()
stream.close()

标签： pythonpython-sounddeviceaws-transcribe

解决方案

我现在相信这BadRequestException是指没有正确编码帧而不是音频数据错误。我发现您的代码存在一些问题：

  1. 您需要以特殊方式对标头/正文进行编码：https ://docs.aws.amazon.com/transcribe/latest/dg/event-stream.html

  2. 您需要对发送的缓冲区非常了解。音频需要是 16 位/无符号 (int)/little-endian（请参见此处）。现在您只是将浮点数（您的麦克风数据浮点数吗？？）转换为 16 位友好值，但将其存储在缓冲区中，缓冲区的位大小由系统确定（实际上可能是 32 位或 64 位），并且您正在使用JSON 字符串编码器。之后不太可能是正确的格式。基本上，您需要一个缓冲区库，它可以让您使用指定的位大小（16）和字节序值（小）设置 int。这是我的飞镖代码，例如：

for (var i=0; i<audioChunk.length; i++) {
  messageBytes.setInt16(offset, audioChunk[i], Endian.little);
  offset += 2;
}
  1. 当您使用上面提到的 EventStream 标头编码时，您必须同样小心那些 32 位长度的整数是 BIG endian。同样的规则适用。您需要以允许您指定位大小和字节序的方式写入字节缓冲区。

最好的方法是编写 AWS 响应所需的解码函数，然后对编码帧进行解码，看看结果是否相同。对 audo 使用测试数据，例如 [-32000, -100, 0, 200 31000] 或类似的东西，这样您就可以测试字节顺序等都是正确的。

推荐阅读