@pompeii-labs/audio 0.1.5 → 0.2.0

package/dist/voice.d.mts

@@ -1,7 +1,5 @@
 import { A as AudioFormat } from './index-o4B-ThOL.mjs';
 import { DeepgramClient, LiveSchema } from '@deepgram/sdk';
-import { ElevenLabsClient } from '@elevenlabs/elevenlabs-js';
-import { StreamTextToSpeechRequest } from '@elevenlabs/elevenlabs-js/api/index.js';
 import { HumeClient } from 'hume';
 import OpenAI from 'openai';
 
@@ -58,6 +56,7 @@ declare class MagmaFlow {
     private generatingAudio;
     private currentRequestId;
     private audioBuffer;
+    private lastChunk;
     private config;
     constructor(args: MagmaFlowArgs);
     inputAudio(audio: Buffer): void;
@@ -120,22 +119,44 @@ declare class DeepgramTTS extends MagmaFlowTextToSpeech {
     reset(): void;
 }
 
-declare enum ElevenVoice {
+declare enum ElevenLabsVoice {
     chris = "iP95p4xoKVk53GoZ742B",
     josh = "TxGEqnHWrfWFTfGW9XjX",
     rachel = "21m00Tcm4TlvDq8ikWAM",
     laura = "FGY2WhTYpPnrIDTdsKH5",
     felicity = "aTbnroHRGIomiKpqAQR8"
 }
-type ElevenLabsConfig = Omit<StreamTextToSpeechRequest, 'outputFormat' | 'text' | 'modelId'>;
+type StreamSpeechConfig = {
+    text: string;
+    model_id?: string;
+    language_code?: string;
+    voice_settings?: {
+        stability?: number;
+        use_speaker_boost?: boolean;
+        similarity_boost?: number;
+        style?: number;
+        speed?: number;
+    };
+    pronunciation_dictionary_locators?: {
+        pronunciation_dictionary_id: string;
+        version_id?: string;
+    }[];
+    seed?: number;
+    next_text?: string;
+    previous_request_ids?: string[];
+    next_request_ids?: string[];
+    apply_text_normalization?: 'auto' | 'on' | 'off';
+    apply_language_text_normalization?: boolean;
+};
+type ElevenLabsConfig = Omit<StreamSpeechConfig, 'text' | 'model_id'>;
 type ElevenLabsTTSArgs = {
-    client?: ElevenLabsClient;
     model: string;
-    voice: ElevenVoice | string;
+    voice: ElevenLabsVoice | string;
     config?: ElevenLabsConfig;
+    apiKey?: string;
 };
 declare class ElevenLabsTTS extends MagmaFlowTextToSpeech {
-    private client;
+    private apiKey;
     private model;
     private voice;
     private config;
@@ -172,4 +193,4 @@ declare class WhisperTTS extends MagmaFlowTextToSpeech {
 
 declare function splitTextIntoChunks(text: string, targetLength?: number): string[];
 
-export { DeepgramLanguage, DeepgramModel, DeepgramSTT, type DeepgramSTTArgs, DeepgramTTS, type DeepgramTTSArgs, type DeepgramWord, ElevenLabsTTS, type ElevenLabsTTSArgs, ElevenVoice, HumeTTS, type HumeTTSArgs, MagmaFlow, type MagmaFlowArgs, type MagmaFlowSTTOutput, MagmaFlowSpeechToText, MagmaFlowTextToSpeech, type Turn, WhisperTTS, type WhisperTTSArgs, splitTextIntoChunks };
+export { DeepgramLanguage, DeepgramModel, DeepgramSTT, type DeepgramSTTArgs, DeepgramTTS, type DeepgramTTSArgs, type DeepgramWord, ElevenLabsTTS, type ElevenLabsTTSArgs, ElevenLabsVoice, HumeTTS, type HumeTTSArgs, MagmaFlow, type MagmaFlowArgs, type MagmaFlowSTTOutput, MagmaFlowSpeechToText, MagmaFlowTextToSpeech, type Turn, WhisperTTS, type WhisperTTSArgs, splitTextIntoChunks };

package/dist/voice.d.ts

@@ -1,7 +1,5 @@
 import { A as AudioFormat } from './index-o4B-ThOL.js';
 import { DeepgramClient, LiveSchema } from '@deepgram/sdk';
-import { ElevenLabsClient } from '@elevenlabs/elevenlabs-js';
-import { StreamTextToSpeechRequest } from '@elevenlabs/elevenlabs-js/api/index.js';
 import { HumeClient } from 'hume';
 import OpenAI from 'openai';
 
@@ -58,6 +56,7 @@ declare class MagmaFlow {
     private generatingAudio;
     private currentRequestId;
     private audioBuffer;
+    private lastChunk;
     private config;
     constructor(args: MagmaFlowArgs);
     inputAudio(audio: Buffer): void;
@@ -120,22 +119,44 @@ declare class DeepgramTTS extends MagmaFlowTextToSpeech {
     reset(): void;
 }
 
-declare enum ElevenVoice {
+declare enum ElevenLabsVoice {
     chris = "iP95p4xoKVk53GoZ742B",
     josh = "TxGEqnHWrfWFTfGW9XjX",
     rachel = "21m00Tcm4TlvDq8ikWAM",
     laura = "FGY2WhTYpPnrIDTdsKH5",
     felicity = "aTbnroHRGIomiKpqAQR8"
 }
-type ElevenLabsConfig = Omit<StreamTextToSpeechRequest, 'outputFormat' | 'text' | 'modelId'>;
+type StreamSpeechConfig = {
+    text: string;
+    model_id?: string;
+    language_code?: string;
+    voice_settings?: {
+        stability?: number;
+        use_speaker_boost?: boolean;
+        similarity_boost?: number;
+        style?: number;
+        speed?: number;
+    };
+    pronunciation_dictionary_locators?: {
+        pronunciation_dictionary_id: string;
+        version_id?: string;
+    }[];
+    seed?: number;
+    next_text?: string;
+    previous_request_ids?: string[];
+    next_request_ids?: string[];
+    apply_text_normalization?: 'auto' | 'on' | 'off';
+    apply_language_text_normalization?: boolean;
+};
+type ElevenLabsConfig = Omit<StreamSpeechConfig, 'text' | 'model_id'>;
 type ElevenLabsTTSArgs = {
-    client?: ElevenLabsClient;
     model: string;
-    voice: ElevenVoice | string;
+    voice: ElevenLabsVoice | string;
     config?: ElevenLabsConfig;
+    apiKey?: string;
 };
 declare class ElevenLabsTTS extends MagmaFlowTextToSpeech {
-    private client;
+    private apiKey;
     private model;
     private voice;
     private config;
@@ -172,4 +193,4 @@ declare class WhisperTTS extends MagmaFlowTextToSpeech {
 
 declare function splitTextIntoChunks(text: string, targetLength?: number): string[];
 
-export { DeepgramLanguage, DeepgramModel, DeepgramSTT, type DeepgramSTTArgs, DeepgramTTS, type DeepgramTTSArgs, type DeepgramWord, ElevenLabsTTS, type ElevenLabsTTSArgs, ElevenVoice, HumeTTS, type HumeTTSArgs, MagmaFlow, type MagmaFlowArgs, type MagmaFlowSTTOutput, MagmaFlowSpeechToText, MagmaFlowTextToSpeech, type Turn, WhisperTTS, type WhisperTTSArgs, splitTextIntoChunks };
+export { DeepgramLanguage, DeepgramModel, DeepgramSTT, type DeepgramSTTArgs, DeepgramTTS, type DeepgramTTSArgs, type DeepgramWord, ElevenLabsTTS, type ElevenLabsTTSArgs, ElevenLabsVoice, HumeTTS, type HumeTTSArgs, MagmaFlow, type MagmaFlowArgs, type MagmaFlowSTTOutput, MagmaFlowSpeechToText, MagmaFlowTextToSpeech, type Turn, WhisperTTS, type WhisperTTSArgs, splitTextIntoChunks };

package/dist/voice.js

@@ -1,7 +1,6 @@
 'use strict';
 
 var sdk = require('@deepgram/sdk');
-var elevenlabsJs = require('@elevenlabs/elevenlabs-js');
 var hume = require('hume');
 var OpenAI = require('openai');
 
@@ -429,6 +428,7 @@ var MagmaFlow = class {
   generatingAudio = false;
   currentRequestId = null;
   audioBuffer = [];
+  lastChunk = null;
   config = {
     pauseDurationMs: 500,
     sentenceChunkLength: 50
@@ -446,9 +446,8 @@ var MagmaFlow = class {
         return;
       }
       if (!audio) {
-        const lastChunk = this.audioBuffer[this.audioBuffer.length - 1];
-        if (lastChunk) {
-          const lastChunkSamples = bufferToInt16Array(lastChunk);
+        if (this.lastChunk) {
+          const lastChunkSamples = bufferToInt16Array(this.lastChunk);
           const lastSampleValue = lastChunkSamples[lastChunkSamples.length - 1];
           this.audioBuffer.push(
             Buffer.from(
@@ -462,10 +461,15 @@ var MagmaFlow = class {
         }
         this.sendAudio();
         this.generatingAudio = false;
+        this.lastChunk = null;
         this.generateAudio();
         return;
       }
       this.audioBuffer.push(audio);
+      this.lastChunk = audio;
+      if (this.audioBuffer.reduce((acc, curr) => acc + curr.length, 0) % (2 * this.outputFormat.channels) === 0) {
+        this.sendAudio();
+      }
     };
     this.stt.onOutput = args.onTranscription;
     this.stt.onSpeechDetected = args.onSpeechDetected;
@@ -505,13 +509,13 @@ var MagmaFlow = class {
   sendAudio() {
     if (this.audioBuffer.length === 0) return;
     const concatenatedBuffer = Buffer.concat(this.audioBuffer);
+    this.audioBuffer = [];
     const resampledPCM = resamplePcm(
       bufferToInt16Array(concatenatedBuffer),
       uniformSampleRate,
       this.outputFormat.sampleRate
     );
     const encodedAudio = encodePcm(resampledPCM, this.outputFormat.encoding);
-    this.audioBuffer = [];
     try {
       this.onAudioOutput(encodedAudio);
     } catch (error) {
@@ -642,7 +646,7 @@ var DeepgramSTT = class extends MagmaFlowSpeechToText {
     }
     const text = this.turnBuffer.map((turn) => turn.text).join(" ");
     let turns = void 0;
-    if (this.turnBuffer.every((turn) => !!turn.speaker)) {
+    if (this.turnBuffer.every((turn) => turn.speaker !== void 0 && turn.speaker !== null)) {
       turns = this.turnBuffer.reduce((acc, turn) => {
         if (acc.at(-1)?.speaker === turn.speaker) {
           acc.at(-1).text += turn.text;
@@ -760,24 +764,24 @@ var DeepgramTTS = class extends MagmaFlowTextToSpeech {
   reset() {
   }
 };
-var ElevenVoice = /* @__PURE__ */ ((ElevenVoice2) => {
-  ElevenVoice2["chris"] = "iP95p4xoKVk53GoZ742B";
-  ElevenVoice2["josh"] = "TxGEqnHWrfWFTfGW9XjX";
-  ElevenVoice2["rachel"] = "21m00Tcm4TlvDq8ikWAM";
-  ElevenVoice2["laura"] = "FGY2WhTYpPnrIDTdsKH5";
-  ElevenVoice2["felicity"] = "aTbnroHRGIomiKpqAQR8";
-  return ElevenVoice2;
-})(ElevenVoice || {});
+
+// src/voice/textToSpeech/elevenlabs.ts
+var ElevenLabsVoice = /* @__PURE__ */ ((ElevenLabsVoice2) => {
+  ElevenLabsVoice2["chris"] = "iP95p4xoKVk53GoZ742B";
+  ElevenLabsVoice2["josh"] = "TxGEqnHWrfWFTfGW9XjX";
+  ElevenLabsVoice2["rachel"] = "21m00Tcm4TlvDq8ikWAM";
+  ElevenLabsVoice2["laura"] = "FGY2WhTYpPnrIDTdsKH5";
+  ElevenLabsVoice2["felicity"] = "aTbnroHRGIomiKpqAQR8";
+  return ElevenLabsVoice2;
+})(ElevenLabsVoice || {});
 var ElevenLabsTTS = class extends MagmaFlowTextToSpeech {
-  client;
+  apiKey;
   model;
   voice;
   config;
   constructor(args) {
     super();
-    this.client = args.client ?? new elevenlabsJs.ElevenLabsClient({
-      apiKey: process.env.ELEVENLABS_API_KEY
-    });
+    this.apiKey = args.apiKey ?? process.env.ELEVENLABS_API_KEY;
     this.model = args.model;
     this.voice = args.voice;
     this.config = args.config ?? {};
@@ -789,14 +793,28 @@ var ElevenLabsTTS = class extends MagmaFlowTextToSpeech {
       return;
     }
     const textToSend = text.replaceAll(/([A-Z])-([A-Z])/g, "$1 - $2").replaceAll(/([0-9])-([0-9])/g, "$1 - $2").replaceAll(/(-\s*[A-Z])\s+([A-Z]\s*-)/g, "$1 - $2").replaceAll(/(-\s*[0-9])\s+([0-9]\s*-)/g, "$1 - $2");
-    this.client.textToSpeech.stream(this.voice, {
-      text: textToSend,
-      outputFormat: "pcm_48000",
-      modelId: this.model,
-      ...this.config
-    }).then(async (stream) => {
-      for await (const chunk of stream) {
-        this.onOutput(chunk, requestId);
+    fetch(
+      `https://api.elevenlabs.io/v1/text-to-speech/${this.voice}/stream?output_format=pcm_48000`,
+      {
+        method: "POST",
+        headers: {
+          "Content-Type": "application/json",
+          "xi-api-key": this.apiKey
+        },
+        body: JSON.stringify({
+          text: textToSend,
+          model_id: this.model,
+          ...this.config
+        })
+      }
+    ).then(async (response) => {
+      const reader = response.body?.getReader();
+      if (!reader) return;
+      new TextDecoder();
+      while (true) {
+        const { done, value } = await reader.read();
+        if (done) break;
+        this.onOutput(Buffer.from(value), requestId);
       }
       this.onOutput(null, requestId);
       console.log("[ElevenLabs] Finished:", textToSend);
@@ -882,7 +900,7 @@ exports.DeepgramModel = DeepgramModel;
 exports.DeepgramSTT = DeepgramSTT;
 exports.DeepgramTTS = DeepgramTTS;
 exports.ElevenLabsTTS = ElevenLabsTTS;
-exports.ElevenVoice = ElevenVoice;
+exports.ElevenLabsVoice = ElevenLabsVoice;
 exports.HumeTTS = HumeTTS;
 exports.MagmaFlow = MagmaFlow;
 exports.MagmaFlowSpeechToText = MagmaFlowSpeechToText;

package/dist/voice.mjs

@@ -1,5 +1,4 @@
 import { DeepgramClient, LiveTranscriptionEvents } from '@deepgram/sdk';
-import { ElevenLabsClient } from '@elevenlabs/elevenlabs-js';
 import { HumeClient } from 'hume';
 import OpenAI from 'openai';
 
@@ -423,6 +422,7 @@ var MagmaFlow = class {
   generatingAudio = false;
   currentRequestId = null;
   audioBuffer = [];
+  lastChunk = null;
   config = {
     pauseDurationMs: 500,
     sentenceChunkLength: 50
@@ -440,9 +440,8 @@ var MagmaFlow = class {
         return;
       }
       if (!audio) {
-        const lastChunk = this.audioBuffer[this.audioBuffer.length - 1];
-        if (lastChunk) {
-          const lastChunkSamples = bufferToInt16Array(lastChunk);
+        if (this.lastChunk) {
+          const lastChunkSamples = bufferToInt16Array(this.lastChunk);
           const lastSampleValue = lastChunkSamples[lastChunkSamples.length - 1];
           this.audioBuffer.push(
             Buffer.from(
@@ -456,10 +455,15 @@ var MagmaFlow = class {
         }
         this.sendAudio();
         this.generatingAudio = false;
+        this.lastChunk = null;
         this.generateAudio();
         return;
       }
       this.audioBuffer.push(audio);
+      this.lastChunk = audio;
+      if (this.audioBuffer.reduce((acc, curr) => acc + curr.length, 0) % (2 * this.outputFormat.channels) === 0) {
+        this.sendAudio();
+      }
     };
     this.stt.onOutput = args.onTranscription;
     this.stt.onSpeechDetected = args.onSpeechDetected;
@@ -499,13 +503,13 @@ var MagmaFlow = class {
   sendAudio() {
     if (this.audioBuffer.length === 0) return;
     const concatenatedBuffer = Buffer.concat(this.audioBuffer);
+    this.audioBuffer = [];
     const resampledPCM = resamplePcm(
       bufferToInt16Array(concatenatedBuffer),
       uniformSampleRate,
       this.outputFormat.sampleRate
     );
     const encodedAudio = encodePcm(resampledPCM, this.outputFormat.encoding);
-    this.audioBuffer = [];
     try {
       this.onAudioOutput(encodedAudio);
     } catch (error) {
@@ -636,7 +640,7 @@ var DeepgramSTT = class extends MagmaFlowSpeechToText {
     }
     const text = this.turnBuffer.map((turn) => turn.text).join(" ");
     let turns = void 0;
-    if (this.turnBuffer.every((turn) => !!turn.speaker)) {
+    if (this.turnBuffer.every((turn) => turn.speaker !== void 0 && turn.speaker !== null)) {
       turns = this.turnBuffer.reduce((acc, turn) => {
         if (acc.at(-1)?.speaker === turn.speaker) {
           acc.at(-1).text += turn.text;
@@ -754,24 +758,24 @@ var DeepgramTTS = class extends MagmaFlowTextToSpeech {
   reset() {
   }
 };
-var ElevenVoice = /* @__PURE__ */ ((ElevenVoice2) => {
-  ElevenVoice2["chris"] = "iP95p4xoKVk53GoZ742B";
-  ElevenVoice2["josh"] = "TxGEqnHWrfWFTfGW9XjX";
-  ElevenVoice2["rachel"] = "21m00Tcm4TlvDq8ikWAM";
-  ElevenVoice2["laura"] = "FGY2WhTYpPnrIDTdsKH5";
-  ElevenVoice2["felicity"] = "aTbnroHRGIomiKpqAQR8";
-  return ElevenVoice2;
-})(ElevenVoice || {});
+
+// src/voice/textToSpeech/elevenlabs.ts
+var ElevenLabsVoice = /* @__PURE__ */ ((ElevenLabsVoice2) => {
+  ElevenLabsVoice2["chris"] = "iP95p4xoKVk53GoZ742B";
+  ElevenLabsVoice2["josh"] = "TxGEqnHWrfWFTfGW9XjX";
+  ElevenLabsVoice2["rachel"] = "21m00Tcm4TlvDq8ikWAM";
+  ElevenLabsVoice2["laura"] = "FGY2WhTYpPnrIDTdsKH5";
+  ElevenLabsVoice2["felicity"] = "aTbnroHRGIomiKpqAQR8";
+  return ElevenLabsVoice2;
+})(ElevenLabsVoice || {});
 var ElevenLabsTTS = class extends MagmaFlowTextToSpeech {
-  client;
+  apiKey;
   model;
   voice;
   config;
   constructor(args) {
     super();
-    this.client = args.client ?? new ElevenLabsClient({
-      apiKey: process.env.ELEVENLABS_API_KEY
-    });
+    this.apiKey = args.apiKey ?? process.env.ELEVENLABS_API_KEY;
     this.model = args.model;
     this.voice = args.voice;
     this.config = args.config ?? {};
@@ -783,14 +787,28 @@ var ElevenLabsTTS = class extends MagmaFlowTextToSpeech {
       return;
     }
     const textToSend = text.replaceAll(/([A-Z])-([A-Z])/g, "$1 - $2").replaceAll(/([0-9])-([0-9])/g, "$1 - $2").replaceAll(/(-\s*[A-Z])\s+([A-Z]\s*-)/g, "$1 - $2").replaceAll(/(-\s*[0-9])\s+([0-9]\s*-)/g, "$1 - $2");
-    this.client.textToSpeech.stream(this.voice, {
-      text: textToSend,
-      outputFormat: "pcm_48000",
-      modelId: this.model,
-      ...this.config
-    }).then(async (stream) => {
-      for await (const chunk of stream) {
-        this.onOutput(chunk, requestId);
+    fetch(
+      `https://api.elevenlabs.io/v1/text-to-speech/${this.voice}/stream?output_format=pcm_48000`,
+      {
+        method: "POST",
+        headers: {
+          "Content-Type": "application/json",
+          "xi-api-key": this.apiKey
+        },
+        body: JSON.stringify({
+          text: textToSend,
+          model_id: this.model,
+          ...this.config
+        })
+      }
+    ).then(async (response) => {
+      const reader = response.body?.getReader();
+      if (!reader) return;
+      new TextDecoder();
+      while (true) {
+        const { done, value } = await reader.read();
+        if (done) break;
+        this.onOutput(Buffer.from(value), requestId);
       }
       this.onOutput(null, requestId);
       console.log("[ElevenLabs] Finished:", textToSend);
@@ -871,4 +889,4 @@ var WhisperTTS = class extends MagmaFlowTextToSpeech {
   }
 };
 
-export { DeepgramLanguage, DeepgramModel, DeepgramSTT, DeepgramTTS, ElevenLabsTTS, ElevenVoice, HumeTTS, MagmaFlow, MagmaFlowSpeechToText, MagmaFlowTextToSpeech, WhisperTTS, splitTextIntoChunks };
+export { DeepgramLanguage, DeepgramModel, DeepgramSTT, DeepgramTTS, ElevenLabsTTS, ElevenLabsVoice, HumeTTS, MagmaFlow, MagmaFlowSpeechToText, MagmaFlowTextToSpeech, WhisperTTS, splitTextIntoChunks };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pompeii-labs/audio",
-  "version": "0.1.5",
+  "version": "0.2.0",
   "description": "The Audio SDK from Pompeii Labs",
   "keywords": [
     "Pompeii",
@@ -42,7 +42,6 @@
   },
   "dependencies": {
     "@deepgram/sdk": "4.2.0",
-    "@elevenlabs/elevenlabs-js": "2.2.0",
     "hume": "0.11.1",
     "openai": "4.86.2"
   },