Speech-to-Text API

The speech recognition endpoint accepts an audio file and returns text using a speech recognition model running on the device. This example uses the SenseVoice model for AI Pyramid.

Preparation

Before calling the endpoint, install the speech recognition model and tools required by the example. Refer to the model library for platform support.

1. Use the apt package manager to install the SenseVoice model package.

apt install llm-model-sense-voice-small-10s-ax650

2. Install the ffmpeg tool.

apt install ffmpeg

3. After installation is complete, restart the OpenAI service to make the new model take effect.

systemctl restart llm-openai-api

Example Program

On the PC side, use the OpenAI API to pass an audio file to implement the speech-to-text function. Before running the example program, modify the IP part of base_url below to the actual IP address of the device.

from openai import OpenAI
client = OpenAI(
    api_key="sk-",
    base_url="http://192.168.20.186:8000/v1"
)

audio_file = open("speech.mp3", "rb")
transcript = client.audio.transcriptions.create(
  model="sense-voice-small-10s-ax650",
  file=audio_file
)

print(transcript)

Request Parameters

Response Example

Transcription(text=' Thank you. Thank you everybody. All right everybody go ahead and have a seat. How\'s everybody doing today? .....',
logprobs=None, task='transcribe', language='en', duration=334.234, segments=12, sample_rate=16000, channels=1, bit_depth=16)