AWS Transcribe Streaming WebSocket Disconnects Immediately

[brightversion1]

Environment information

Service: AWS Transcribe (Streaming WebSocket API)

Frontend: React + WebSocket + MediaRecorder/AudioContext

Backend: AWS Lambda (Node.js) generating presigned URL using @aws-sdk/signature-v4

Region: eu-west-2

Audio format: PCM16, 16 kHz, mono

Describe the bug

When connecting to AWS Transcribe using a presigned WebSocket URL, the connection closes immediately with no error message from the service.

Even when sending empty or valid PCM16 audio chunks, the socket disconnects right after onopen.
The client never receives transcription events.

Reproduction steps

1. Generate Presigned URL in Lambda

const request = new HttpRequest({
  protocol: "wss:",
  hostname: `transcribestreaming.${region}.amazonaws.com`,
  path: "/stream-transcription-websocket",
  method: "GET",
  query: {
    "language-code": "en-US",
    "media-encoding": "pcm",
    "sample-rate": "16000"
  },
  headers: {
    host: `transcribestreaming.${region}.amazonaws.com`,
  },
});

const signer = new SignatureV4({
  credentials: defaultProvider(),
  region,
  service: "transcribe",
  sha256: Sha256,
});

const signedRequest = await signer.presign(request, { expiresIn: 300 });
const url = `wss://${signedRequest.hostname}:8443${signedRequest.path}?${queryStr}`;

2. Connect From Frontend

const socket = new WebSocket(wsUrl);

socket.onopen = () => {
  console.log("✅ Connected to Transcribe");
};

socket.onmessage = (event) => {
  console.log("Message:", event.data);
};

socket.onerror = (event) => {
  console.error("WebSocket error:", event);
};

socket.onclose = (event) => {
  console.log("🔴 Disconnected", event.code, event.reason, event.wasClean);
};

3. Send Audio

processor.onaudioprocess = (e) => {
  if (socket.readyState !== 1) return;

  const float32Array = e.inputBuffer.getChannelData(0);
  const int16Array = new Int16Array(float32Array.length);
  for (let i = 0; i < float32Array.length; i++) {
    int16Array[i] = Math.max(-32768, Math.min(32767, Math.floor(float32Array[i] * 32768)));
  }

  const audioEvent = {
    headers: {
      ":content-type": { type: "string", value: "application/octet-stream" },
      ":event-type": { type: "string", value: "AudioEvent" },
      ":message-type": { type: "string", value: "event" },
    },
    body: new Uint8Array(int16Array.buffer),
  };

  socket.send(marshaller.marshall(audioEvent));
};

---

⚠️ Observed Behavior

• WebSocket connects (onopen fires ✅).
• Immediately disconnects, with no transcription events.
• onclose shows no reason (event.reason empty).
• Adding silent/empty audio does not change behavior.

---

✅ Expected Behavior

• WebSocket should remain open until closed by client.
• Transcribe should return transcription events or at least an error message when audio is invalid.
• Empty audio should result in clean closure, not silent disconnect.

---

❓ Questions for AWS Team

1. Is there an issue with how the presigned URL is generated (path, port :8443, or query params)?
2. Does Transcribe WebSocket require a special initial handshake message before sending audio chunks?
3. What are the exact requirements for the AudioEvent payload format (PCM16, marshalling, headers)?
4. Why does the connection close without returning an error or BadRequestException?

---

Would you like me to also include a Minimal Working Example repo structure (with Lambda + React frontend code) in this issue so AWS support can reproduce it in one go?

[pahud]

Hi @brightversion1,

Thank you for the detailed issue report. However, this issue is not related to Amplify Backend - it's about integrating with the AWS Transcribe Streaming WebSocket API directly.

This repository is for issues with the Amplify Backend framework itself (e.g., defineBackend, defineFunction, defineData, etc.). Your issue is about AWS service integration and should be directed to:

1. AWS Support - For service-specific questions about Transcribe Streaming API
2. aws-sdk-js repository - For SDK-related issues with @aws-sdk/client-transcribe-streaming
3. AWS re:Post - For community help with AWS service integrations

Quick Diagnosis

Looking at your code, I see several potential issues:

1. Service Name: You're using service: "transcribe" but it should likely be "transcribe-streaming"

2. URL Construction: You're manually reconstructing the URL which bypasses the signed query parameters:

   // ❌ This loses the signed parameters
   const url = `wss://${signedRequest.hostname}:8443${signedRequest.path}?${queryStr}`;

3. Protocol Mismatch: Signing with protocol: "wss:" may not work - you typically sign with https: then convert to wss:

Recommended Approach

Instead of manually creating presigned URLs, use the official AWS SDK which handles all WebSocket complexity:

import { TranscribeStreamingClient, StartStreamTranscriptionCommand } from "@aws-sdk/client-transcribe-streaming";

const client = new TranscribeStreamingClient({ region: "eu-west-2" });
// SDK handles WebSocket connection, event stream encoding, etc.

If you need browser-based access, consider using API Gateway WebSocket with a Lambda proxy that connects to Transcribe using the SDK.

Next Steps

I'll set pending-response on this issue as it's not related to Amplify Backend. Please:

• Open a support case with AWS for Transcribe-specific questions
• Check the AWS Transcribe Streaming documentation
• Consider posting on AWS re:Post for community assistance

If you believe this is actually an Amplify Backend issue, please clarify how you're using Amplify Backend in your setup.