@ducci/jarvis 1.0.86 → 1.0.87

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ducci/jarvis",
-  "version": "1.0.86",
+  "version": "1.0.87",
   "description": "A fully automated agent system that lives on a server.",
   "main": "./src/index.js",
   "type": "module",

package/src/server/fish-audio.js

@@ -3,6 +3,12 @@
  */
 
 import { createClient } from './provider.js';
+import { exec } from 'child_process';
+import { promisify } from 'util';
+import { writeFile, readFile, unlink } from 'fs/promises';
+import { tmpdir } from 'os';
+import { join } from 'path';
+const execAsync = promisify(exec);
 
 // System prompt for TTS summary generation.
 // fish.audio s1 emotion tags: (emotion) at the START of a sentence only — applies to the whole sentence.
@@ -97,10 +103,24 @@ export async function textToSpeech(text, config) {
 export async function speechToText(audioBuffer, config) {
   const { fishAudioApiKey } = config;
 
+  // Telegram voice messages are OGG/Opus — fish.audio ASR doesn't support Opus.
+  // Convert to WAV first via ffmpeg.
+  const id = `jarvis-stt-${Date.now()}`;
+  const inPath = join(tmpdir(), `${id}.ogg`);
+  const outPath = join(tmpdir(), `${id}.wav`);
+  let wavBuffer;
+  try {
+    await writeFile(inPath, audioBuffer);
+    await execAsync(`ffmpeg -y -i "${inPath}" -ar 16000 -ac 1 "${outPath}"`);
+    wavBuffer = await readFile(outPath);
+  } finally {
+    unlink(inPath).catch(() => {});
+    unlink(outPath).catch(() => {});
+  }
+
   const formData = new FormData();
-  const blob = new Blob([audioBuffer], { type: 'audio/ogg' });
-  formData.append('audio', blob, 'voice.ogg');
-  formData.append('ignore_timestamps', 'true');
+  const blob = new Blob([wavBuffer], { type: 'audio/wav' });
+  formData.append('audio', blob, 'voice.wav');
 
   const response = await fetch('https://api.fish.audio/v1/asr', {
     method: 'POST',