fixing eleven labs api

This commit is contained in:
akdeb 2026-02-17 16:43:33 +07:00
parent 0407b4a561
commit fe5a0c8cce
3 changed files with 343 additions and 281 deletions

View file

@ -33,11 +33,11 @@ volatile bool sleepRequested = false;
*/ */
#ifdef DEV_MODE #ifdef DEV_MODE
const char *ws_server = "192.168.1.155"; const char *ws_server = "192.168.1.121";
const uint16_t ws_port = 8000; const uint16_t ws_port = 8000;
const char *ws_path = "/"; const char *ws_path = "/";
// Backend server details // Backend server details
const char *backend_server = "192.168.1.155"; const char *backend_server = "192.168.1.121";
const uint16_t backend_port = 3000; const uint16_t backend_port = 3000;
#elif defined(PROD_MODE) #elif defined(PROD_MODE)

View file

@ -4,7 +4,7 @@ import type {
WebSocket as WSWebSocket, WebSocket as WSWebSocket,
WebSocketServer as _WebSocketServer, WebSocketServer as _WebSocketServer,
} from "npm:@types/ws"; } from "npm:@types/ws";
import { authenticateUser, elevenLabsApiKey } from "./utils.ts"; import { authenticateUser } from "./utils.ts";
import { import {
createFirstMessage, createFirstMessage,
createSystemPrompt, createSystemPrompt,
@ -59,7 +59,7 @@ wss.on("connection", async (ws: WSWebSocket, payload: IPayload) => {
ws.send( ws.send(
JSON.stringify({ JSON.stringify({
type: "auth", type: "auth",
volume_control: user.device?.volume ?? 20, volume_control: user.device?.volume ?? 100,
is_ota: user.device?.is_ota ?? false, is_ota: user.device?.is_ota ?? false,
is_reset: user.device?.is_reset ?? false, is_reset: user.device?.is_reset ?? false,
pitch_factor: user.personality?.pitch_factor ?? 1, pitch_factor: user.personality?.pitch_factor ?? 1,
@ -92,20 +92,7 @@ wss.on("connection", async (ws: WSWebSocket, payload: IPayload) => {
await connectToGrok(providerArgs); await connectToGrok(providerArgs);
break; break;
case "elevenlabs": case "elevenlabs":
const agentId = user.personality?.oai_voice ?? ""; await connectToElevenLabs(providerArgs);
if (!elevenLabsApiKey) {
throw new Error("ELEVENLABS_API_KEY environment variable is required");
}
await connectToElevenLabs(
ws,
payload,
connectionPcmFile,
agentId,
elevenLabsApiKey,
closeHandler,
);
break; break;
case "hume": case "hume":
await connectToHume(providerArgs); await connectToHume(providerArgs);

View file

@ -7,9 +7,13 @@ import {
type IncomingSocketEvent, type IncomingSocketEvent,
type DisconnectionDetails type DisconnectionDetails
} from "npm:@elevenlabs/client"; } from "npm:@elevenlabs/client";
import { addConversation, getDeviceInfo } from "../supabase.ts"; import { addConversation, getDeviceInfo } from "../supabase.ts";
import { createOpusPacketizer, isDev } from "../utils.ts"; import {
createOpusPacketizer,
isDev,
elevenLabsApiKey,
SAMPLE_RATE,
} from "../utils.ts";
// Calculate audio level for debugging // Calculate audio level for debugging
function calculateAudioLevel(audioData: any): number { function calculateAudioLevel(audioData: any): number {
@ -26,40 +30,83 @@ function calculateAudioLevel(audioData: any): number {
return Math.round(sum / samples.length); return Math.round(sum / samples.length);
} }
export const connectToElevenLabs = async ( // Resample mono PCM16 little-endian audio with linear interpolation.
ws: WebSocket, function resamplePcm16Mono(
payload: IPayload, inputBytes: Buffer,
connectionPcmFile: Deno.FsFile | null, fromRate: number,
agentId: string, toRate: number,
apiKey: string, ): Buffer {
closeHandler: () => Promise<void>, if (fromRate === toRate || inputBytes.length === 0) {
) => { return inputBytes;
console.log(apiKey, agentId); }
const { user, supabase } = payload;
const inputSamples = inputBytes.length / 2;
const outputSamples = Math.max(1, Math.floor((inputSamples * toRate) / fromRate));
const output = Buffer.alloc(outputSamples * 2);
for (let i = 0; i < outputSamples; i++) {
const sourcePos = (i * fromRate) / toRate;
const leftIndex = Math.floor(sourcePos);
const rightIndex = Math.min(leftIndex + 1, inputSamples - 1);
const frac = sourcePos - leftIndex;
const left = inputBytes.readInt16LE(leftIndex * 2);
const right = inputBytes.readInt16LE(rightIndex * 2);
const sample = Math.round(left + (right - left) * frac);
output.writeInt16LE(sample, i * 2);
}
return output;
}
export const connectToElevenLabs = async ({
ws,
payload,
connectionPcmFile,
firstMessage,
closeHandler,
}: ProviderArgs) => {
const agentId = payload.user.personality?.voice?.config?.config_id ??
payload.user.personality?.oai_voice;
const apiKey = elevenLabsApiKey;
if (!agentId || !apiKey) {
throw new Error("Agent ID or API key is missing");
}
const { user, supabase } = payload;
const opus = createOpusPacketizer((packet) => ws.send(packet)); const opus = createOpusPacketizer((packet) => ws.send(packet));
// Queue messages until ElevenLabs connection is ready // Queue messages until ElevenLabs connection is ready.
const messageQueue: RawData[] = []; const messageQueue: Array<{ data: RawData; isBinary: boolean }> = [];
let isElevenLabsConnected = false; let isElevenLabsConnected = false;
let elevenLabsConnection: WebSocketConnection | null = null; let elevenLabsConnection: WebSocketConnection | null = null;
let hasResponseStarted = false; let hasResponseStarted = false;
let elevenInputSampleRate = 16000;
let elevenOutputSampleRate = 16000;
// Handle messages from ESP32 client // Handle messages from ESP32 ws client.
const handleClientMessage = async (data: any, isBinary: boolean) => { const handleClientMessage = async (data: any, isBinary: boolean) => {
try { try {
if (isBinary) { if (isBinary) {
const base64Data = data.toString("base64");
if (isDev && connectionPcmFile) { if (isDev && connectionPcmFile) {
await connectionPcmFile.write(data); await connectionPcmFile.write(data);
} }
// Send audio to ElevenLabs using their client // Send audio to ElevenLabs using the expected input sample rate.
if (isElevenLabsConnected && elevenLabsConnection) { if (isElevenLabsConnected && elevenLabsConnection) {
// Check if audio contains actual speech (simple volume check) const sourceBuffer = Buffer.from(data);
const pcmForEleven = resamplePcm16Mono(
sourceBuffer,
SAMPLE_RATE,
elevenInputSampleRate,
);
const base64Data = pcmForEleven.toString("base64");
const audioLevel = calculateAudioLevel(data); const audioLevel = calculateAudioLevel(data);
console.log(`Sending audio chunk to ElevenLabs: raw=${data.length} bytes, base64=${base64Data.length} chars, level=${audioLevel}`); console.log(
`Sending audio chunk to ElevenLabs: raw=${data.length} bytes, resampled=${pcmForEleven.length} bytes, inRate=${SAMPLE_RATE}, elevenInRate=${elevenInputSampleRate}, level=${audioLevel}`,
);
try { try {
elevenLabsConnection.sendMessage({ elevenLabsConnection.sendMessage({
@ -69,7 +116,9 @@ export const connectToElevenLabs = async (
console.error("Error sending audio to ElevenLabs:", error); console.error("Error sending audio to ElevenLabs:", error);
} }
} else { } else {
console.log(`Cannot send audio - ElevenLabs connected: ${isElevenLabsConnected}, connection exists: ${!!elevenLabsConnection}`); console.log(
`Cannot send audio - ElevenLabs connected: ${isElevenLabsConnected}, connection exists: ${!!elevenLabsConnection}`,
);
} }
} else { } else {
const message = JSON.parse(data.toString("utf-8")); const message = JSON.parse(data.toString("utf-8"));
@ -80,7 +129,7 @@ export const connectToElevenLabs = async (
console.log("Interrupt detected"); console.log("Interrupt detected");
if (elevenLabsConnection) { if (elevenLabsConnection) {
elevenLabsConnection.sendMessage({ elevenLabsConnection.sendMessage({
type: "user_activity" type: "user_activity",
}); });
} }
break; break;
@ -99,39 +148,70 @@ export const connectToElevenLabs = async (
} }
}; };
// Register handlers immediately so early ESP32 audio is not dropped while ElevenLabs connects.
ws.on("message", (data: any, isBinary: boolean) => {
if (!isElevenLabsConnected) {
messageQueue.push({ data, isBinary });
} else {
handleClientMessage(data, isBinary);
}
});
ws.on("error", (error: any) => {
console.error("ESP32 WebSocket error:", error);
opus.close();
elevenLabsConnection?.close();
});
ws.on("close", async (code: number, reason: string) => {
console.log(`ESP32 WebSocket closed with code ${code}, reason: ${reason}`);
await closeHandler();
opus.close();
elevenLabsConnection?.close();
if (isDev && connectionPcmFile) {
connectionPcmFile.close();
console.log("Closed debug audio file.");
}
});
try { try {
// For server-side usage, we need to get a signed URL first // For server-side usage, we need to get a signed URL first.
const signedUrlResponse = await fetch( const signedUrlResponse = await fetch(
`https://api.elevenlabs.io/v1/convai/conversation/get-signed-url?agent_id=${agentId}`, `https://api.elevenlabs.io/v1/convai/conversation/get-signed-url?agent_id=${agentId}`,
{ {
headers: { headers: {
'xi-api-key': apiKey, "xi-api-key": apiKey,
},
}, },
}
); );
if (!signedUrlResponse.ok) { if (!signedUrlResponse.ok) {
throw new Error(`Failed to get signed URL: ${signedUrlResponse.status} ${signedUrlResponse.statusText}`); throw new Error(
`Failed to get signed URL: ${signedUrlResponse.status} ${signedUrlResponse.statusText}`,
);
} }
const { signed_url } = await signedUrlResponse.json(); const { signed_url } = await signedUrlResponse.json();
// Use default audio formats (let ElevenLabs auto-detect)
const modifiedSignedUrl = signed_url;
// Create ElevenLabs connection using signed URL for server-side usage
const sessionConfig: SessionConfig = { const sessionConfig: SessionConfig = {
signedUrl: modifiedSignedUrl, signedUrl: signed_url,
connectionType: "websocket", connectionType: "websocket",
}; };
elevenLabsConnection = await WebSocketConnection.create(sessionConfig); elevenLabsConnection = await WebSocketConnection.create(sessionConfig);
elevenInputSampleRate = elevenLabsConnection.inputFormat.sampleRate;
elevenOutputSampleRate = elevenLabsConnection.outputFormat.sampleRate;
console.log("Connected to ElevenLabs successfully!"); console.log("Connected to ElevenLabs successfully!");
console.log(
`ElevenLabs formats: input=${elevenLabsConnection.inputFormat.format}_${elevenInputSampleRate}, output=${elevenLabsConnection.outputFormat.format}_${elevenOutputSampleRate}`,
);
isElevenLabsConnected = true; isElevenLabsConnected = true;
console.log(`ElevenLabs connection ready - conversation_initiation_metadata already processed by SDK`); console.log(
"ElevenLabs connection ready - conversation_initiation_metadata already processed by SDK",
);
// Set up ElevenLabs event handlers
elevenLabsConnection.onMessage(async (event: IncomingSocketEvent) => { elevenLabsConnection.onMessage(async (event: IncomingSocketEvent) => {
console.log("ElevenLabs message type:", event); console.log("ElevenLabs message type:", event);
@ -141,40 +221,48 @@ export const connectToElevenLabs = async (
break; break;
case "ping": case "ping":
// Handle ping messages - send pong response
console.log("Received ping from ElevenLabs, sending pong"); console.log("Received ping from ElevenLabs, sending pong");
if (event.ping_event?.event_id) { if (event.ping_event?.event_id) {
elevenLabsConnection.sendMessage({ elevenLabsConnection?.sendMessage({
type: "pong", type: "pong",
event_id: event.ping_event.event_id event_id: event.ping_event.event_id,
}); });
} }
break; break;
case "audio": case "audio":
if (event.audio_event?.audio_base_64) { if (event.audio_event?.audio_base_64) {
// Send RESPONSE.CREATED before first audio chunk
if (!hasResponseStarted) { if (!hasResponseStarted) {
console.log("Sending RESPONSE.CREATED to ESP32 (agent audio starting)"); console.log(
"Sending RESPONSE.CREATED to ESP32 (agent audio starting)",
);
opus.reset(); opus.reset();
ws.send(JSON.stringify({ ws.send(JSON.stringify({
type: "server", type: "server",
msg: "RESPONSE.CREATED" msg: "RESPONSE.CREATED",
})); }));
hasResponseStarted = true; hasResponseStarted = true;
} }
const audioBuffer = Buffer.from(event.audio_event.audio_base_64, "base64"); const audioBuffer = Buffer.from(
console.log(`Received audio from ElevenLabs: ${audioBuffer.length} bytes`); event.audio_event.audio_base_64,
"base64",
// Use Opus packetizer to encode and send audio );
opus.push(audioBuffer); const pcmForEsp32 = resamplePcm16Mono(
audioBuffer,
elevenOutputSampleRate,
SAMPLE_RATE,
);
opus.push(pcmForEsp32);
} }
break; break;
case "user_transcript": case "user_transcript":
if (event.user_transcription_event?.user_transcript) { if (event.user_transcription_event?.user_transcript) {
console.log("User transcript:", event.user_transcription_event.user_transcript); console.log(
"User transcript:",
event.user_transcription_event.user_transcript,
);
addConversation( addConversation(
supabase, supabase,
"user", "user",
@ -183,11 +271,12 @@ export const connectToElevenLabs = async (
); );
if (!hasResponseStarted) { if (!hasResponseStarted) {
console.log("Sending RESPONSE.CREATED to ESP32 (agent audio starting)"); console.log(
opus.reset(); "Sending RESPONSE.CREATED to ESP32 (agent audio starting)",
);
ws.send(JSON.stringify({ ws.send(JSON.stringify({
type: "server", type: "server",
msg: "RESPONSE.CREATED" msg: "RESPONSE.CREATED",
})); }));
hasResponseStarted = true; hasResponseStarted = true;
} }
@ -204,12 +293,9 @@ export const connectToElevenLabs = async (
user, user,
); );
// Flush any remaining audio before sending complete
opus.flush(true);
// Send response complete with device info like OpenAI does
console.log("Sending RESPONSE.COMPLETE to ESP32"); console.log("Sending RESPONSE.COMPLETE to ESP32");
hasResponseStarted = false; // Reset for next response opus.flush(true);
hasResponseStarted = false;
try { try {
const device = await getDeviceInfo(supabase, user.user_id); const device = await getDeviceInfo(supabase, user.user_id);
ws.send(JSON.stringify({ ws.send(JSON.stringify({
@ -228,16 +314,21 @@ export const connectToElevenLabs = async (
break; break;
case "vad_score": case "vad_score":
// Voice Activity Detection score - can be used for debugging
if (event.vad_score_event?.vad_score) { if (event.vad_score_event?.vad_score) {
console.log("VAD score:", event.vad_score_event.vad_score); console.log("VAD score:", event.vad_score_event.vad_score);
} }
break; break;
case "internal_tentative_agent_response": case "internal_tentative_agent_response":
// Tentative response while agent is thinking if (
if (event.tentative_agent_response_internal_event?.tentative_agent_response) { event.tentative_agent_response_internal_event
console.log("Tentative response:", event.tentative_agent_response_internal_event.tentative_agent_response); ?.tentative_agent_response
) {
console.log(
"Tentative response:",
event.tentative_agent_response_internal_event
.tentative_agent_response,
);
} }
break; break;
@ -245,7 +336,7 @@ export const connectToElevenLabs = async (
console.log("ElevenLabs conversation ended"); console.log("ElevenLabs conversation ended");
ws.send(JSON.stringify({ ws.send(JSON.stringify({
type: "server", type: "server",
msg: "SESSION.END" msg: "SESSION.END",
})); }));
break; break;
@ -256,46 +347,30 @@ export const connectToElevenLabs = async (
elevenLabsConnection.onDisconnect((details: DisconnectionDetails) => { elevenLabsConnection.onDisconnect((details: DisconnectionDetails) => {
console.log("ElevenLabs connection closed:", details.reason); console.log("ElevenLabs connection closed:", details.reason);
opus.close();
ws.close(); ws.close();
}); });
// Process queued messages // Match OpenAI/Gemini flow by triggering the first assistant turn immediately.
if (firstMessage?.trim()) {
console.log("Sending initial user_message to ElevenLabs to start first turn");
elevenLabsConnection.sendMessage({
type: "user_message",
text: firstMessage,
});
}
// Process queued messages.
while (messageQueue.length > 0) { while (messageQueue.length > 0) {
const queuedMessage = messageQueue.shift(); const queuedMessage = messageQueue.shift();
if (queuedMessage) { if (queuedMessage) {
handleClientMessage(queuedMessage, false); handleClientMessage(queuedMessage.data, queuedMessage.isBinary);
} }
} }
// Set up ESP32 WebSocket handlers
ws.on("message", (data: any, isBinary: boolean) => {
if (!isElevenLabsConnected) {
messageQueue.push(data);
} else {
handleClientMessage(data, isBinary);
}
});
ws.on("error", (error: any) => {
console.error("ESP32 WebSocket error:", error);
elevenLabsConnection?.close();
});
ws.on("close", async (code: number, reason: string) => {
console.log(`ESP32 WebSocket closed with code ${code}, reason: ${reason}`);
await closeHandler();
opus.close();
elevenLabsConnection?.close();
if (isDev && connectionPcmFile) {
connectionPcmFile.close();
console.log("Closed debug audio file.");
}
});
} catch (error) { } catch (error) {
console.error("Failed to connect to ElevenLabs:", error); console.error("Failed to connect to ElevenLabs:", error);
opus.close();
// Send more specific error information
let errorMessage = "RESPONSE.ERROR"; let errorMessage = "RESPONSE.ERROR";
if (error instanceof Error) { if (error instanceof Error) {
console.error("Error details:", error.message); console.error("Error details:", error.message);
@ -306,7 +381,7 @@ export const connectToElevenLabs = async (
ws.send(JSON.stringify({ ws.send(JSON.stringify({
type: "server", type: "server",
msg: errorMessage msg: errorMessage,
})); }));
} }
}; };