能否通过Python将预录音频文件传入Google Assistant SDK并构建API？

Let's break down your questions one by one, with practical Python-focused solutions tailored to your use case.

1. Can I directly pass an audio file as input to the Google Assistant SDK? #

Not exactly "directly" in the sense of passing a file path, but you can feed pre-recorded audio data to the SDK—so long as the audio meets strict format requirements:

• 16kHz sampling rate
• Mono (single channel)
• Uncompressed PCM (typically saved as a .raw file)

The SDK expects audio input as a byte stream or raw audio bytes, not a file reference. So you’ll need to read your pre-recorded file into memory and pass those bytes to the SDK’s input methods.

2. Can I trigger a fixed pre-recorded command via a button/API endpoint? #

Absolutely! This is totally feasible with the Google Assistant SDK, and Python is perfect for building the API endpoint you need. Here’s a step-by-step implementation plan plus a working code example:

Prerequisites #

First, get your environment set up:

• Enable the Google Assistant API in your Google Cloud Project, then create and download service account credentials (save as credentials.json).
• Install required Python packages:

  pip install google-assistant-sdk[samples] flask google-auth
  

• Convert your pre-recorded command audio to the correct format (if it’s not already). Use ffmpeg for this quick conversion:

  ffmpeg -i your-command.mp3 -acodec pcm_s16le -ac 1 -ar 16000 command.raw
  

Python API Endpoint Example (Using Flask) #

This code creates a simple POST endpoint that triggers the Assistant with your pre-recorded audio when called—exactly what you need for a button-triggered workflow:

from flask import Flask, jsonify
import google.auth
from google.assistant.library import Assistant
from google.assistant.library.event import EventType
import os

# Initialize Flask app
app = Flask(__name__)

# Load pre-recorded audio once at startup (avoids reloading on every request)
PRE_RECORDED_AUDIO_PATH = "command.raw"
if not os.path.exists(PRE_RECORDED_AUDIO_PATH):
    raise FileNotFoundError(f"Audio file {PRE_RECORDED_AUDIO_PATH} not found. Check your conversion step!")

with open(PRE_RECORDED_AUDIO_PATH, "rb") as f:
    PRE_RECORDED_AUDIO = f.read()

def fetch_assistant_response():
    # Load Google credentials (set the GOOGLE_APPLICATION_CREDENTIALS env var to your credentials.json path)
    credentials, project_id = google.auth.default(scopes=[
        "https://www.googleapis.com/auth/assistant-sdk-prototype"
    ])
    
    with Assistant(credentials, project_id) as assistant:
        assistant.start_conversation()
        # Send the pre-recorded audio bytes to the Assistant
        assistant.send_audio(PRE_RECORDED_AUDIO)
        
        # Listen for and return the Assistant's response
        for event in assistant.start():
            if event.type == EventType.ON_RESPONSE_RECEIVED:
                response_text = event.args.get('text', 'No text response received')
                assistant.stop_conversation()
                return response_text
            elif event.type == EventType.ON_CONVERSATION_TERMINATED:
                assistant.stop_conversation()
                return "Conversation ended without a response"

@app.route('/trigger-command', methods=['POST'])
def trigger_command():
    try:
        response = fetch_assistant_response()
        return jsonify({
            "status": "success",
            "assistant_response": response
        })
    except Exception as e:
        return jsonify({
            "status": "error",
            "message": str(e)
        }), 500

if __name__ == '__main__':
    # Run the API endpoint (accessible at http://localhost:5000/trigger-command)
    app.run(host='0.0.0.0', port=5000, debug=True)

Key Tips for Production #

• Audio Format Check: Double-check your .raw file matches the 16kHz mono PCM requirement—incorrect format will make the Assistant ignore your input.
• Authentication: Ensure the GOOGLE_APPLICATION_CREDENTIALS environment variable points to your credentials.json file before running the app.
• Performance Optimization: For high-traffic use, reuse the Assistant instance instead of creating a new one per request (use a singleton pattern to reduce latency).
• Error Handling: Expand the basic error handling to cover cases like missing credentials, audio read failures, or SDK timeouts.

内容的提问来源于stack exchange，提问作者joke4me