用 Amazon Nova Sonic + WebRTC 搭建实时语音交互直播应用

观众麦克风 → WebRTC Audio Track → Signaling → Media Server
    → Nova Sonic (语音识别 + 语义理解 + 语音合成)
    → 合成音频回推 WebRTC → 观众扬声器

import boto3
import json
import asyncio
from amazon_kinesis_video_streaming_webrtc_client import (
    KvsWebRTCClient,
    SignalingClient,
)

# ---------- 1. 创建信令通道 ----------
kvs = boto3.client("kinesisvideo", region_name="us-east-1")
channel_name = "live-voice-demo"

# 确保通道存在（生产环境应提前创建）
try:
    kvs.create_signaling_channel(
        ChannelName=channel_name,
        ChannelType="SINGLE_MASTER",
    )
except kvs.exceptions.ResourceAlreadyExistsException:
    pass  # 通道已存在，继续

# ---------- 2. Master 端接入 WebRTC ----------
signaling_client = SignalingClient(
    channel_name=channel_name,
    region="us-east-1",
    role="MASTER",
    # 使用 IAM Role 或硬编码凭证（示例用后者，生产请用 Role）
    access_key="AKIAIOSFODNN7EXAMPLE",
    secret_key="wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY",
)

webrtc_client = KvsWebRTCClient(signaling_client=signaling_client)

# ---------- 3. 音频帧回调 → Nova Sonic ----------
nova_client = boto3.client("nova-sonic", region_name="us-east-1")
session_id = "demo-session-001"

# 启动 Nova Sonic 流式会话（伪接口，展示调用模式）
nova_session = nova_client.start_streaming_session(
    SessionId=session_id,
    InputFormat="pcm_16khz_mono",   # WebRTC 默认音频格式
    OutputFormat="pcm_16khz_mono",
    VoiceId="ai-host-v1",            # 选择合成音色
)

async def on_audio_frame(frame_bytes: bytes):
    """WebRTC 收到 Viewer 音频帧后的回调"""
    # 喂给 Nova Sonic
    nova_client.send_audio_chunk(
        SessionId=session_id,
        AudioData=frame_bytes,
    )

    # 从 Nova Sonic 拿合成音频（流式，非阻塞）
    response = nova_client.receive_audio_chunk(SessionId=session_id)
    if response and response.get("AudioData"):
        # 回推给 WebRTC Viewer
        webrtc_client.send_audio_frame(response["AudioData"])

# ---------- 4. 启动 ----------
async def main():
    await signaling_client.connect()
    await webrtc_client.start(on_audio_frame=on_audio_frame)
    print("Master 端已启动，等待 Viewer 连接…")
    # 保持运行
    await asyncio.Event().wait()

asyncio.run(main())

// ---------- Viewer 端：浏览器 JavaScript ----------
const channelName = "live-voice-demo";
const region = "us-east-1";

// 1. 通过 KVS Signaling API 获取 ICE Server 配置和 SDP offer
//    （实际项目中这一步通常由后端 API 代理，避免凭证暴露到前端）
const signalingUrl = `https://kvssignaling.${region}.amazonaws.com`;
const resp = await fetch(`/api/webrtc-signaling?channel=${channelName}`);
const { sdpOffer, iceServers } = await resp.json();

// 2. 创建 PeerConnection，加入 ICE servers
const pc = new RTCPeerConnection({ iceServers });

// 3. 添加本地麦克风 Track
const stream = await navigator.mediaDevices.getUserMedia({ audio: true });
stream.getAudioTracks().forEach(track => pc.addTrack(track, stream));

// 4. 接收远端（Master/Nova Sonic）音频
pc.ontrack = (event) => {
    const remoteAudio = document.getElementById("remoteAudio");
    remoteAudio.srcObject = event.streams[0];
    remoteAudio.play();
};

// 5. SDP 交换
await pc.setRemoteDescription(new RTCSessionDescription(sdpOffer));
const answer = await pc.createAnswer();
await pc.setLocalDescription(answer);

// 把 answer 回传给信令服务
await fetch(`/api/webrtc-answer?channel=${channelName}`, {
    method: "POST",
    body: JSON.stringify({ sdpAnswer: answer.sdp }),
});