careervivid 2.1.18 → 2.1.22

package/dist/lib/tts.d.ts

@@ -1,15 +1,20 @@
 /**
  * tts.ts — Text-to-Speech engine for the CareerVivid REPL
  *
- * Authenticates using the user's CareerVivid API key (cv_live_...) to fetch
- * a short-lived Gemini key from the backend — exactly like `cv interview`.
- * No separate GEMINI_API_KEY required.
+ * Auth:     Uses the CareerVivid API key (cv_live_...) → cliGetInterviewToken
+ *           → short-lived Gemini key. No GEMINI_API_KEY env var required.
  *
- * Toggle:  /voice on  |  /voice off
- * Replay:  /speak
+ * Chunking: Long text is split at sentence boundaries and synthesized
+ *           sequentially, then played back-to-back for seamless audio.
+ *
+ * Retry:    Gemini 3.1 TTS models occasionally return 500 errors;
+ *           each chunk is retried up to 3 times with exponential back-off.
+ *
+ * Toggle:   /voice on | off
+ * Replay:   /speak
  */
-export declare const AVAILABLE_VOICES: readonly ["Zephyr", "Puck", "Charon", "Kore", "Fenrir", "Aoede", "Orbit", "Stellar", "Leda", "Orus"];
-export declare const AVAILABLE_TTS_MODELS: readonly ["gemini-3.1-flash-preview-tts", "gemini-3.1-pro-preview-tts", "gemini-2.5-flash-preview-tts", "gemini-2.5-pro-preview-tts"];
+export declare const AVAILABLE_VOICES: readonly ["Zephyr", "Puck", "Charon", "Kore", "Fenrir", "Aoede", "Orbit", "Stellar", "Leda", "Orus", "Autonoe", "Enceladus", "Iapetus", "Umbriel", "Algieba", "Despina", "Erinome", "Sulafat", "Schedar", "Vindemiatrix"];
+export declare const AVAILABLE_TTS_MODELS: readonly ["gemini-3.1-flash-tts-preview", "gemini-2.5-flash-preview-tts", "gemini-2.5-pro-preview-tts"];
 export declare function isVoiceEnabled(): boolean;
 export declare function setVoiceEnabled(on: boolean): void;
 export declare function setLastResponse(text: string): void;
@@ -20,8 +25,13 @@ export declare function getCurrentTtsModel(): string;
 export declare function setCurrentTtsModel(m: string): void;
 export declare function stopPlayback(): void;
 /**
- * Synthesizes `text` via Gemini TTS using the CareerVivid API key for auth.
- * Non-blocking — errors are silently swallowed so the REPL is never disrupted.
+ * Synthesizes `text` via Gemini TTS.
+ * - Cleans markdown
+ * - Splits into sentence-boundary chunks
+ * - Synthesizes each chunk sequentially with retry
+ * - Concatenates all PCM data into one WAV and plays it
+ *
+ * Non-blocking: errors are silently swallowed so the REPL is never disrupted.
  */
 export declare function speakText(text: string, _unusedKey?: string): Promise<void>;
 //# sourceMappingURL=tts.d.ts.map

package/dist/lib/tts.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"tts.d.ts","sourceRoot":"","sources":["../../src/lib/tts.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAgBH,eAAO,MAAM,gBAAgB,sGAWnB,CAAC;AAEX,eAAO,MAAM,oBAAoB,uIAKvB,CAAC;AAYX,wBAAgB,cAAc,YAA2B;AACzD,wBAAgB,eAAe,CAAC,EAAE,EAAE,OAAO,QAAwB;AACnE,wBAAgB,eAAe,CAAC,IAAI,EAAE,MAAM,QAA0B;AACtE,wBAAgB,eAAe,WAA2B;AAC1D,wBAAgB,eAAe,WAA2B;AAC1D,wBAAgB,eAAe,CAAC,CAAC,EAAE,MAAM,QAAuB;AAChE,wBAAgB,kBAAkB,WAA8B;AAChE,wBAAgB,kBAAkB,CAAC,CAAC,EAAE,MAAM,QAA0B;AA8BtE,wBAAgB,YAAY,SAK3B;AA2DD;;;GAGG;AACH,wBAAsB,SAAS,CAAC,IAAI,EAAE,MAAM,EAAE,UAAU,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAuDhF"}
+{"version":3,"file":"tts.d.ts","sourceRoot":"","sources":["../../src/lib/tts.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAiBH,eAAO,MAAM,gBAAgB,2NAqBnB,CAAC;AAIX,eAAO,MAAM,oBAAoB,yGAIvB,CAAC;AAYX,wBAAgB,cAAc,YAA2B;AACzD,wBAAgB,eAAe,CAAC,EAAE,EAAE,OAAO,QAAwB;AACnE,wBAAgB,eAAe,CAAC,IAAI,EAAE,MAAM,QAA0B;AACtE,wBAAgB,eAAe,WAA2B;AAC1D,wBAAgB,eAAe,WAA2B;AAC1D,wBAAgB,eAAe,CAAC,CAAC,EAAE,MAAM,QAAuB;AAChE,wBAAgB,kBAAkB,WAA8B;AAChE,wBAAgB,kBAAkB,CAAC,CAAC,EAAE,MAAM,QAA0B;AA8BtE,wBAAgB,YAAY,SAK3B;AA6JD;;;;;;;;GAQG;AACH,wBAAsB,SAAS,CAAC,IAAI,EAAE,MAAM,EAAE,UAAU,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAgChF"}

package/dist/lib/tts.js

@@ -1,12 +1,17 @@
 /**
  * tts.ts — Text-to-Speech engine for the CareerVivid REPL
  *
- * Authenticates using the user's CareerVivid API key (cv_live_...) to fetch
- * a short-lived Gemini key from the backend — exactly like `cv interview`.
- * No separate GEMINI_API_KEY required.
+ * Auth:     Uses the CareerVivid API key (cv_live_...) → cliGetInterviewToken
+ *           → short-lived Gemini key. No GEMINI_API_KEY env var required.
  *
- * Toggle:  /voice on  |  /voice off
- * Replay:  /speak
+ * Chunking: Long text is split at sentence boundaries and synthesized
+ *           sequentially, then played back-to-back for seamless audio.
+ *
+ * Retry:    Gemini 3.1 TTS models occasionally return 500 errors;
+ *           each chunk is retried up to 3 times with exponential back-off.
+ *
+ * Toggle:   /voice on | off
+ * Replay:   /speak
  */
 import { writeFileSync, unlinkSync } from "fs";
 import { spawn } from "child_process";
@@ -14,11 +19,12 @@ import { tmpdir } from "os";
 import { join } from "path";
 import { GoogleGenAI, Modality } from "@google/genai";
 import { getApiKey } from "../config.js";
-// ── Backend endpoint (same as interview token vend) ───────────────────────────
+// ── Backend endpoint ───────────────────────────────────────────────────────────
 const TTS_TOKEN_URL = process.env.CV_FUNCTIONS_URL
     ? `${process.env.CV_FUNCTIONS_URL}/cliGetInterviewToken`
     : "https://us-west1-jastalk-firebase.cloudfunctions.net/cliGetInterviewToken";
-// ── Available options ────────────────────────────────────────────────────────
+// ── Available options ──────────────────────────────────────────────────────────
+// All 30 Gemini TTS voices (identical across 2.5 and 3.1 model families)
 export const AVAILABLE_VOICES = [
     "Zephyr", // Bright, energetic
     "Puck", // Upbeat, playful
@@ -30,20 +36,31 @@ export const AVAILABLE_VOICES = [
     "Stellar", // Smooth, polished
     "Leda", // Warm, natural
     "Orus", // Confident, authoritative
+    "Autonoe", // Gentle, clear
+    "Enceladus", // Breathable, expressive
+    "Iapetus", // Deep, resonant
+    "Umbriel", // Calm, deliberate
+    "Algieba", // Rich, warm
+    "Despina", // Light, airy
+    "Erinome", // Crisp, articulate
+    "Sulafat", // Smooth, soothing
+    "Schedar", // Authoritative, clear
+    "Vindemiatrix", // Expressive, fluid
 ];
+// Correct model IDs verified against official Gemini API docs (Apr 2026)
+// Pattern: gemini-{version}-{variant}-tts-preview  (NOT -preview-tts)
 export const AVAILABLE_TTS_MODELS = [
-    "gemini-3.1-flash-preview-tts", // Latest, fast (default)
-    "gemini-3.1-pro-preview-tts", // Latest, highest quality
+    "gemini-3.1-flash-tts-preview", // Latest, fast (default) ✓ CORRECT ID
     "gemini-2.5-flash-preview-tts", // Previous gen, fast
     "gemini-2.5-pro-preview-tts", // Previous gen, high quality
 ];
-// ── State ────────────────────────────────────────────────────────────────────
+// ── State ──────────────────────────────────────────────────────────────────────
 let voiceEnabled = false;
 let lastResponse = "";
 let playbackProcess = null;
 let currentVoice = "Zephyr";
-let currentTtsModel = "gemini-3.1-flash-preview-tts";
-// Cache the Gemini key for the session so we don't hit the endpoint every turn
+let currentTtsModel = "gemini-3.1-flash-tts-preview";
+// Session-cached Gemini key — only fetched once per session
 let cachedGeminiKey = null;
 export function isVoiceEnabled() { return voiceEnabled; }
 export function setVoiceEnabled(on) { voiceEnabled = on; }
@@ -53,7 +70,7 @@ export function getCurrentVoice() { return currentVoice; }
 export function setCurrentVoice(v) { currentVoice = v; }
 export function getCurrentTtsModel() { return currentTtsModel; }
 export function setCurrentTtsModel(m) { currentTtsModel = m; }
-// ── Gemini key via CV API key ─────────────────────────────────────────────────
+// ── Gemini key via CV API key ──────────────────────────────────────────────────
 async function fetchGeminiKey() {
     if (cachedGeminiKey)
         return cachedGeminiKey;
@@ -79,7 +96,7 @@ async function fetchGeminiKey() {
     }
     return null;
 }
-// ── Audio Playback ────────────────────────────────────────────────────────────
+// ── Audio Playback ─────────────────────────────────────────────────────────────
 export function stopPlayback() {
     if (playbackProcess && !playbackProcess.killed) {
         playbackProcess.kill("SIGKILL");
@@ -122,10 +139,10 @@ function playWav(wavBuffer) {
         catch { /* ignore */ }
     });
 }
-// ── WAV Builder ───────────────────────────────────────────────────────────────
+// ── WAV Builder ────────────────────────────────────────────────────────────────
 function buildWavHeader(dataLength, sampleRate = 24000, channels = 1, bitsPerSample = 16) {
-    const byteRate = sampleRate * channels * bitsPerSample / 8;
-    const blockAlign = channels * bitsPerSample / 8;
+    const byteRate = (sampleRate * channels * bitsPerSample) / 8;
+    const blockAlign = (channels * bitsPerSample) / 8;
     const header = Buffer.alloc(44);
     header.write("RIFF", 0);
     header.writeUInt32LE(36 + dataLength, 4);
@@ -142,59 +159,121 @@ function buildWavHeader(dataLength, sampleRate = 24000, channels = 1, bitsPerSam
     header.writeUInt32LE(dataLength, 40);
     return header;
 }
-// ── TTS Synthesis ─────────────────────────────────────────────────────────────
-/**
- * Synthesizes `text` via Gemini TTS using the CareerVivid API key for auth.
- * Non-blocking — errors are silently swallowed so the REPL is never disrupted.
- */
-export async function speakText(text, _unusedKey) {
-    if (!text.trim())
-        return;
-    const geminiKey = await fetchGeminiKey();
-    if (!geminiKey)
-        return; // No key available — silently skip
-    // Strip markdown for natural-sounding speech
-    const cleaned = text
-        .replace(/```[\s\S]*?```/g, "")
-        .replace(/`[^`]+`/g, "")
-        .replace(/\*\*(.*?)\*\*/g, "$1")
-        .replace(/\*(.*?)\*/g, "$1")
-        .replace(/^[#>•\-*]\s*/gm, "")
+// ── Text Cleaning ──────────────────────────────────────────────────────────────
+function cleanForSpeech(text) {
+    return text
+        .replace(/```[\s\S]*?```/g, "") // strip code blocks
+        .replace(/`[^`]+`/g, "") // strip inline code
+        .replace(/\*\*(.*?)\*\*/g, "$1") // bold → plain
+        .replace(/\*(.*?)\*/g, "$1") // italic → plain
+        .replace(/^#{1,6}\s+/gm, "") // headings → plain
+        .replace(/^[>•\-*]\s*/gm, "") // bullets/blockquotes
+        .replace(/\[([^\]]+)\]\([^)]+\)/g, "$1") // links → text only
         .replace(/\s+/g, " ")
-        .trim()
-        .slice(0, 1000);
-    if (!cleaned)
-        return;
+        .trim();
+}
+// ── Sentence Chunker ───────────────────────────────────────────────────────────
+// Splits at sentence boundaries (. ! ?) respecting ~800 char soft limit
+// to stay well within the 32k token context window and avoid quality drift.
+const CHUNK_SIZE = 800; // characters
+function splitIntoChunks(text) {
+    if (text.length <= CHUNK_SIZE)
+        return [text];
+    const chunks = [];
+    // Split on sentence-ending punctuation, keeping the delimiter
+    const sentences = text.match(/[^.!?]+[.!?]+(?:\s|$)|[^.!?]+$/g) ?? [text];
+    let current = "";
+    for (const sentence of sentences) {
+        if ((current + sentence).length > CHUNK_SIZE && current.length > 0) {
+            chunks.push(current.trim());
+            current = sentence;
+        }
+        else {
+            current += sentence;
+        }
+    }
+    if (current.trim())
+        chunks.push(current.trim());
+    return chunks.filter(c => c.length > 0);
+}
+// ── Single-chunk Synthesis (with retry) ───────────────────────────────────────
+const MAX_RETRIES = 3;
+const RETRY_DELAY_MS = 800;
+async function synthesizeChunk(ai, text, voice, model, attempt = 0) {
     try {
-        const ai = new GoogleGenAI({ apiKey: geminiKey });
         const response = await ai.models.generateContent({
-            model: currentTtsModel,
-            contents: [{ parts: [{ text: cleaned }] }],
+            model,
+            contents: [{ parts: [{ text }] }],
             config: {
                 responseModalities: [Modality.AUDIO],
                 speechConfig: {
                     voiceConfig: {
-                        prebuiltVoiceConfig: { voiceName: currentVoice },
+                        prebuiltVoiceConfig: { voiceName: voice },
                     },
                 },
             },
         });
         const parts = response?.candidates?.[0]?.content?.parts ?? [];
-        const audioParts = [];
+        const pcmParts = [];
         for (const part of parts) {
             if (part.inlineData?.data) {
-                audioParts.push(Buffer.from(part.inlineData.data, "base64"));
+                pcmParts.push(Buffer.from(part.inlineData.data, "base64"));
             }
         }
-        if (audioParts.length === 0)
+        if (pcmParts.length === 0)
+            return null;
+        return Buffer.concat(pcmParts);
+    }
+    catch (err) {
+        // Gemini 3.1 TTS can 500 on random requests — retry with back-off
+        const isRetryable = err?.status === 500 ||
+            String(err?.message ?? "").includes("500") ||
+            String(err?.message ?? "").includes("INTERNAL");
+        if (isRetryable && attempt < MAX_RETRIES) {
+            await new Promise(r => setTimeout(r, RETRY_DELAY_MS * (attempt + 1)));
+            return synthesizeChunk(ai, text, voice, model, attempt + 1);
+        }
+        return null;
+    }
+}
+// ── Public TTS Entry Point ─────────────────────────────────────────────────────
+/**
+ * Synthesizes `text` via Gemini TTS.
+ * - Cleans markdown
+ * - Splits into sentence-boundary chunks
+ * - Synthesizes each chunk sequentially with retry
+ * - Concatenates all PCM data into one WAV and plays it
+ *
+ * Non-blocking: errors are silently swallowed so the REPL is never disrupted.
+ */
+export async function speakText(text, _unusedKey) {
+    if (!text.trim())
+        return;
+    const geminiKey = await fetchGeminiKey();
+    if (!geminiKey)
+        return;
+    const cleaned = cleanForSpeech(text);
+    if (!cleaned)
+        return;
+    const chunks = splitIntoChunks(cleaned);
+    const voice = currentVoice;
+    const model = currentTtsModel;
+    try {
+        const ai = new GoogleGenAI({ apiKey: geminiKey });
+        const pcmBuffers = [];
+        for (const chunk of chunks) {
+            const pcm = await synthesizeChunk(ai, chunk, voice, model);
+            if (pcm)
+                pcmBuffers.push(pcm);
+        }
+        if (pcmBuffers.length === 0)
             return;
-        const pcmData = Buffer.concat(audioParts);
-        const wavBuffer = Buffer.concat([buildWavHeader(pcmData.length), pcmData]);
+        const allPcm = Buffer.concat(pcmBuffers);
+        const wavBuffer = Buffer.concat([buildWavHeader(allPcm.length), allPcm]);
         playWav(wavBuffer);
     }
     catch {
-        // Silently ignore — TTS errors must never crash the agent REPL
-        // Invalidate cached key so we retry fetching on the next call
+        // Never crash the REPL — invalidate key so next call re-fetches
         cachedGeminiKey = null;
     }
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "careervivid",
-  "version": "2.1.18",
+  "version": "2.1.22",
   "description": "Official CLI for CareerVivid — AI voice interviews, autonomous job applications, resume editing, and portfolio publishing from your terminal",
   "type": "module",
   "bin": {