Bad request error

[samueljazzjohn]

Which Deepgram product are you using?

Prerecorded Speech-To-Text.

Details

 with sr.AudioFile(audio_file) as source:
                audio_data = {'buffer': audio_file, 'mimetype': 'audio/wav'}

                response = dg_client.transcription.prerecorded(audio_data,{'model':'whisper'} )
                task = loop.create_task(response)  # Schedule the coroutine to run in the background
                result = loop.run_until_complete(task)  # Execute the coroutine and get the result
                print("text", result)

This code snippet showing some bad request error. I need to get the text transcription from chunked audio datas real time. Is there any other way to do that. Is my code snippet wrong?

If you are making a request to the Deepgram API, what is the full Deepgram URL you are making a request to?

https://api.deepgram.com/v1/listen?model=whisper

If you are making a request to the Deepgram API and have a request ID, please paste it below:

No response

If possible, please attach your code or paste it into the text box.

No response

If possible, please attach an example audio file to reproduce the issue.

No response

[jjmaldonis]

Hey @samueljazzjohn your code looks good and works for me, so it could be two things:

1. The value of audio_file could be incorrect
2. The audio/wav mime type may need to be changed

To figure out if it's the mimetype, take a look at the audio file's extension and look up the mimetype for that audio file. For example, .wav files use a mime type of audio/wav and mp3s use audio/mpeg.

Below is the complete code that works for me (it took a while to run with whisper-large, so run it on Nova if you're testing with English - Nova is cheaper too):

from deepgram import Deepgram
import asyncio
import os


DEEPGRAM_API_KEY = os.environ["DEEPGRAM_API_KEY"]  # Your Deepgram API Key


def transcribe():
    dg_client = Deepgram(DEEPGRAM_API_KEY)
    loop = asyncio.get_event_loop()

    # Download the audio from: https://static.deepgram.com/examples/en_NatGen_Medical_DocDictation.m4a
    with open("./test-audio-files/en_NatGen_Medical_DocDictation.m4a", "rb") as audio_file:
        audio_data = {"buffer": audio_file, "mimetype": "audio/mp4"}
        response = dg_client.transcription.prerecorded(audio_data, {"model": "whisper"})
        task = loop.create_task(
            response
        )  # Schedule the coroutine to run in the background
        result = loop.run_until_complete(
            task
        )  # Execute the coroutine and get the result
        print("text", result)


if __name__ == "__main__":
    transcribe()

[samueljazzjohn]

I am actually using Bytes_IO to handle an audio file, which is a chunk of an audio clip. I am attaching the code snippet below.

            audio, sample_rate = sf.read(file_path)
            chunk_size = int(randum_chuck * sample_rate)

            chunk = audio[temp_chunk_size:temp_chunk_size+chunk_size]
            temp_chunk_size = temp_chunk_size + chunk_size
            bytes_io = io.BytesIO()
            sf.write(bytes_io, chunk, sample_rate, format='wav')
            bytes_io.seek(0)
            audio_queue.put((bytes_io, randum_chuck, i))

Does it word using dg_client.transcription.prerecorded(). I am also attatching the full code snippet for transcribing the text from audio

def recognize_speech_deepgram():
    r = sr.Recognizer()
    loop = asyncio.new_event_loop() 
    asyncio.set_event_loop(loop) 
    loop = asyncio.get_event_loop()
    while True:
        print('recognize, -----------------')
        try:
            audio_file, randum_chuck, i = audio_queue.get()

            with sr.AudioFile(audio_file) as source:
                audio_data = {'buffer': audio_file, 'mimetype': 'audio/wav'}

                try:
                    response = dg_client.transcription.prerecorded(audio_data,{'punctuate': True,'model':'whisper','smart_format': True} )
                    task = loop.create_task(response)  # Schedule the coroutine to run in the background
                    result = loop.run_until_complete(task)  # Execute the coroutine and get the result
                    print("text", result)
                except sr.UnknownValueError:
                    logger.debug("Could not understand audio")
                except sr.RequestError as e:
                    logger.error(e)
        except Exception as e:
            logger.exception(e)
            sleep(0.01)
        except:
            logger.error(traceback.format_exc())
            sleep(0.01)

What I am actually doing is retrieving audio from a .wav file, randomly dividing it into chunks, and placing each audio chunk into an audio queue. Then, I am transcribing each audio chunk using Deepgram. Can I achieve this using the current method, or should I consider using a different approach?

[jjmaldonis]

I can't quite tell how audio_file is getting created, but it sounds like it's an io.BytesIO object. An io.BytesIO object cannot be passed to the Deepgram SDK. Instead, you'll need to pass an io.BufferedReader object. Luckily, the conversation is very simple! You can convert it via io.BufferedReader(audio_data).

Below is a synchronous example that demonstrates how this works:

from deepgram import Deepgram
import asyncio
import io
import os


DEEPGRAM_API_KEY = os.environ["DEEPGRAM_API_KEY"]  # Your Deepgram API Key


async def transcribe():
    deepgram = Deepgram(DEEPGRAM_API_KEY)

    # Download the audio from: https://traffic.megaphone.fm/GLT4320773090.mp3
    # `one_each.mp3` is an audio segement starting at 1213 seconds and going for 12 seconds
    # `longer.mp3` is an audio segment starting at 1207 seconds and going for 120 seconds
    with open("./test-audio-files/longer.mp3", "rb") as f:
        content = f.read()
        f.seek(0)
        bio = io.BytesIO(content[:len(content)//2])
        first = io.BufferedReader(bio)
        bio = io.BytesIO(content[len(content)//2:])
        second = io.BufferedReader(bio)

        audio_data = {"buffer": first, "mimetype": "audio/mpeg"}
        result = await deepgram.transcription.prerecorded(audio_data, {"model": "nova", "diarize": "true"})
        words = result["results"]["channels"][0]["alternatives"][0]["words"]
        for word in words:
            print(word)

        input("Press any key to transcribe the second half of the audio...")
        audio_data = {"buffer": second, "mimetype": "audio/mpeg"}
        result = await deepgram.transcription.prerecorded(audio_data, {"model": "nova", "diarize": "true"})
        words = result["results"]["channels"][0]["alternatives"][0]["words"]
        for word in words:
            print(word)


if __name__ == "__main__":
    asyncio.get_event_loop().run_until_complete(transcribe())

[samueljazzjohn]

Thank you so much for your help. Finally I got the data in Json format is there any way to get the transcript in text format directly

{'word': 'system', 'start': 3.04, 'end': 3.4399998, 'confidence': 0.9995117, 'speaker': 0, 'speaker_confidence': 1.0}
{'word': 'would', 'start': 3.4399998, 'end': 3.6, 'confidence': 0.9980469, 'speaker': 0, 'speaker_confidence': 1.0}
{'word': 'work', 'start': 3.6, 'end': 4.0, 'confidence': 0.9995117, 'speaker': 0, 'speaker_confidence': 1.0}
{'word': 'is', 'start': 4.0, 'end': 4.4, 'confidence': 0.9223633, 'speaker': 0, 'speaker_confidence': 1.0}
{'word': 'that', 'start': 4.4, 'end': 4.9, 'confidence': 0.99658203, 'speaker': 0, 'speaker_confidence': 1.0}
{'word': 'it', 'start': 6.08, 'end': 6.16, 'confidence': 0.9116211, 'speaker': 0, 'speaker_confidence': 1.0}
{'word': 'would', 'start': 6.16, 'end': 6.3999996, 'confidence': 0.9946289, 'speaker': 0, 'speaker_confidence': 1.0}
{'word': 'offer', 'start': 6.3999996, 'end': 6.72, 'confidence': 1.0, 'speaker': 0, 'speaker_confidence': 1.0}
{'word': 'you', 'start': 6.72, 'end': 7.04, 'confidence': 0.99902344, 'speaker': 0, 'speaker_confidence': 1.0}
{'word': 'a', 'start': 7.04, 'end': 7.44, 'confidence': 0.8798828, 'speaker': 0, 'speaker_confidence': 1.0}
{'word': 'simple', 'start': 7.44, 'end': 7.94, 'confidence': 0.9995117, 'speaker': 0, 'speaker_confidence': 1.0}```

[jjmaldonis]

Great glad it worked!

If you parse a different part of the response, you can find the transcript in text format. So rather than using

words = result["results"]["channels"][0]["alternatives"][0]["words"]

you can use

transcript = result["results"]["channels"][0]["alternatives"][0]["transcript"]

[samueljazzjohn]

Thank you for your response. It helped.

I would like to know that how to get speaker information from this transcription.

[jjmaldonis]

If you set diarize=true, the speaker information will contained in the words part of the response. Please see our documentation here for the details: https://developers.deepgram.com/docs/diarization