用于Google的Speech API的MP3到FLAC

Question

I'm trying to find a simple way to send an MP3 to Google for speech recognition. Currently, I'm using a sub process to call SoX which converts it to a WAV. Then, using SpeechRecognition , it converts it again to FLAC. Ideally, I'd like a more portable (not OS specific) way to decode the MP3 and send it with no intermediate file saving and the like.

Here's what I have currently:

import speech_recognition as sr
import subprocess
import requests

audio = requests.get('http://somesite.com/some.mp3')

with open('/tmp/audio.mp3', 'wb') as file:
    file.write(audio.content)

subprocess.run(['sox', '/tmp/audio.mp3', '/tmp/audio.wav'])

r = sr.Recognizer()
with sr.WavFile('/tmp/audio.wav') as source:
    audio = r.record(source)

result = r.recognize_google(audio)
del r

I've tried directly using the FLAC binaries included in SpeechRecognition , but the output was just static. I'm not too keen on distributing binaries on Git, but I will if that is the only way.

Some important links:

SR's code for speech recognition

SR's code for WAV to FLAC

Edit

I'm considering distributing SoX in a way like the FLAC binaries were, one for each OS, if SoX's license allows it...

Second thought, software licenses are confusing and I don't want to mess with that.

Answer 1

I decided to go with this:

import subprocess
import requests
import shutil
import glob
import json

audio = requests.get('http://somesite.com/some.mp3')
sox = shutil.which('sox') or glob.glob('C:\Program Files*\sox*\sox.exe')[0]
p = subprocess.Popen(sox + ' -t mp3 - -t flac - rate 16k', stdin = subprocess.PIPE, stdout = subprocess.PIPE, shell = True)
stdout, stderr = p.communicate(audio.content)
url = 'http://www.google.com/speech-api/v2/recognize?client=chromium&lang=en-US&key=AIzaSyBOti4mM-6x9WDnZIjIeyEU21OpBXqWBgw'
headers = {'Content-Type': 'audio/x-flac; rate=16000'}
response = requests.post(url, data = stdout, headers = headers).text

result = None
for line in response.split('\n'):
    try:
        result = json.loads(line)['result'][0]['alternative'][0]['transcript']
        break
    except:
        pass

This is more of a middle ground I suppose borrowing some stuff from the SR module. It would require the user to install SoX, but should work on all OS and doesn't have any intermediate files. I have only tested it on Linux however.