@qvac/sdk 0.9.1 → 0.10.0

package/dist/examples/llamacpp-native-tools.js

@@ -61,14 +61,8 @@ try {
     // Consume tool call events
     const toolsTask = (async () => {
         for await (const evt of result.toolCallStream) {
-            if (evt.type === "toolCall") {
-                console.log(`\n\n→ Tool Call Detected: ${evt.call.name}(${JSON.stringify(evt.call.arguments)})`);
-                console.log(`   ID: ${evt.call.id}`);
-            }
-            else if (evt.type === "toolCallError") {
-                console.warn(`\n⚠️  Tool Error: ${evt.error.message}`);
-                console.warn(`   Code: ${evt.error.code}`);
-            }
+            console.log(`\n\n→ Tool Call Detected: ${evt.call.name}(${JSON.stringify(evt.call.arguments)})`);
+            console.log(`   ID: ${evt.call.id}`);
         }
     })();
     await Promise.all([tokensTask, toolsTask]);

package/dist/examples/llamacpp-structured-output.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=llamacpp-structured-output.d.ts.map

package/dist/examples/llamacpp-structured-output.js

@@ -0,0 +1,106 @@
+import { loadModel, completion, unloadModel, QWEN3_600M_INST_Q4, } from "@qvac/sdk";
+// Demonstrates per-request structured output via `responseFormat`.
+// Three modes are exercised against the same model so you can eyeball the
+// difference between free-form text, free-form JSON, and a strict JSON Schema.
+//
+// Usage:
+//   bun run examples/llamacpp-structured-output.ts
+const PERSON_SCHEMA = {
+    type: "object",
+    properties: {
+        name: { type: "string" },
+        age: { type: "integer" },
+        occupation: { type: "string" },
+    },
+    required: ["name", "age", "occupation"],
+    additionalProperties: false,
+};
+const HISTORY = [
+    {
+        role: "system",
+        content: "You extract structured information about people from short bios. /no_think",
+    },
+    {
+        role: "user",
+        content: "Hi, I'm Alice, I'm 30 years old and I work as a data engineer.",
+    },
+];
+// `json_object` mode only enforces that the output is *some* valid JSON object
+// — it doesn't pin the keys. Small models (Qwen3-0.6B in this example) will
+// often emit `{}` because that's the shortest valid completion under the
+// grammar. The same gotcha is documented for OpenAI's `response_format:
+// json_object`. To get useful keys you either need a stronger prompt + a
+// larger model, or — better — switch to `json_schema` (mode 3 below) so the
+// grammar itself forces the keys.
+const JSON_OBJECT_HISTORY = [
+    {
+        role: "system",
+        content: 'You extract structured information about people from short bios and reply ONLY with a JSON object containing "name", "age", and "occupation". /no_think',
+    },
+    {
+        role: "user",
+        content: "Hi, I'm Alice, I'm 30 years old and I work as a data engineer.",
+    },
+];
+// Drains the run via the canonical `events` / `final` surface and returns the
+// content text. `tokenStream` / `text` / `stats` still work but are legacy and
+// will be deprecated, so new code should consume `events` for streaming and
+// await `final.contentText` for the aggregated result.
+async function runToContent(run) {
+    for await (const event of run.events) {
+        if (event.type === "contentDelta")
+            process.stdout.write(event.text);
+    }
+    const final = await run.final;
+    process.stdout.write("\n");
+    return final.contentText;
+}
+try {
+    const modelId = await loadModel({
+        modelSrc: QWEN3_600M_INST_Q4,
+        modelType: "llm",
+        onProgress: (p) => process.stdout.write(`\rLoading: ${p.percentage.toFixed(1)}%`),
+    });
+    console.log(`\nModel loaded: ${modelId}\n`);
+    console.log("--- 1. responseFormat: text (baseline, free-form) ---");
+    await runToContent(completion({
+        modelId,
+        history: HISTORY,
+        stream: true,
+        responseFormat: { type: "text" },
+    }));
+    console.log("\n--- 2. responseFormat: json_object (any valid JSON) ---");
+    const jsonObjectOut = await runToContent(completion({
+        modelId,
+        history: JSON_OBJECT_HISTORY,
+        stream: true,
+        responseFormat: { type: "json_object" },
+    }));
+    console.log("parsed:", JSON.parse(jsonObjectOut.trim()));
+    console.log("\n--- 3. responseFormat: json_schema (strict shape) ---");
+    const jsonSchemaOut = await runToContent(completion({
+        modelId,
+        history: HISTORY,
+        stream: true,
+        responseFormat: {
+            type: "json_schema",
+            json_schema: {
+                name: "person",
+                schema: PERSON_SCHEMA,
+            },
+        },
+    }));
+    // Output is guaranteed schema-valid JSON: object with exactly
+    // {name: string, age: integer, occupation: string}, no extras.
+    const parsed = JSON.parse(jsonSchemaOut.trim());
+    console.log("parsed:", parsed);
+    console.log("schema-valid:", typeof parsed.name === "string" &&
+        Number.isInteger(parsed.age) &&
+        typeof parsed.occupation === "string" &&
+        Object.keys(parsed).sort().join(",") === "age,name,occupation");
+    await unloadModel({ modelId });
+}
+catch (error) {
+    console.error("Error:", error);
+    process.exit(1);
+}

package/dist/examples/llamacpp-tools-pythonic.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=llamacpp-tools-pythonic.d.ts.map

package/dist/examples/llamacpp-tools-pythonic.js

@@ -0,0 +1,69 @@
+import { z } from "zod";
+import { completion, loadModel, unloadModel, } from "@qvac/sdk";
+// LFM2-1.2B-Tool emits Pythonic-style calls: [get_weather(city="Tokyo")].
+const LFM_TOOL_HF = "https://huggingface.co/LiquidAI/LFM2-1.2B-Tool-GGUF/resolve/main/LFM2-1.2B-Tool-Q4_K_M.gguf";
+const weatherSchema = z.object({
+    city: z.string().describe("City name"),
+});
+const horoscopeSchema = z.object({
+    sign: z.string().describe("An astrological sign like Taurus or Aquarius"),
+});
+const tools = [
+    {
+        name: "get_weather",
+        description: "Get current weather for a city",
+        parameters: weatherSchema,
+    },
+    {
+        name: "get_horoscope",
+        description: "Get today's horoscope for an astrological sign",
+        parameters: horoscopeSchema,
+    },
+];
+try {
+    const modelId = await loadModel({
+        modelSrc: LFM_TOOL_HF,
+        modelType: "llm",
+        modelConfig: { ctx_size: 4096, tools: true },
+        onProgress: (progress) => console.log(`Loading: ${progress.percentage.toFixed(1)}%`),
+    });
+    console.log(`✅ Model loaded: ${modelId}`);
+    const history = [
+        {
+            role: "system",
+            content: "You are a helpful assistant that can call tools to look up weather and horoscopes.",
+        },
+        {
+            role: "user",
+            content: "What's the weather in Tokyo and my horoscope for Aquarius?",
+        },
+    ];
+    console.log("\n🤖 Streaming...\n");
+    const result = completion({ modelId, history, stream: true, tools });
+    const tokensTask = (async () => {
+        for await (const token of result.tokenStream) {
+            process.stdout.write(token);
+        }
+    })();
+    const toolsTask = (async () => {
+        for await (const evt of result.toolCallStream) {
+            console.log(`\n→ ${evt.call.name}(${JSON.stringify(evt.call.arguments)})`);
+        }
+    })();
+    await Promise.all([tokensTask, toolsTask]);
+    const toolCalls = await result.toolCalls;
+    console.log("\n\n📋 Final tool calls:");
+    if (toolCalls.length > 0) {
+        for (const call of toolCalls) {
+            console.log(`  - ${call.name}(${JSON.stringify(call.arguments)})`);
+        }
+    }
+    else {
+        console.log("  (none)");
+    }
+    await unloadModel({ modelId });
+}
+catch (error) {
+    console.error("❌ Error:", error);
+    process.exit(1);
+}

package/dist/examples/plugins.js

@@ -46,7 +46,7 @@ process.on("exit", cleanup);
 // Point config before importing the SDK.
 // In a real app, place qvac.config.json (or .js/.ts) in your project root (auto-discovered).
 process.env["QVAC_CONFIG_PATH"] = configPath;
-const { completion, translate, embed, loadModel, unloadModel, LLAMA_3_2_1B_INST_Q4_0, MARIAN_OPUS_EN_IT_Q4_0, GTE_LARGE_FP16, } = await import("@qvac/sdk");
+const { completion, translate, embed, loadModel, unloadModel, LLAMA_3_2_1B_INST_Q4_0, BERGAMOT_EN_ES, GTE_LARGE_FP16, } = await import("@qvac/sdk");
 console.log("1. LLM Completion (llamacpp-completion plugin)");
 try {
     const llmModelId = await loadModel({
@@ -78,12 +78,12 @@ catch (error) {
 console.log("2. Translation (nmtcpp-translation plugin)");
 try {
     const nmtModelId = await loadModel({
-        modelSrc: MARIAN_OPUS_EN_IT_Q4_0,
+        modelSrc: BERGAMOT_EN_ES,
         modelType: "nmt",
         modelConfig: {
-            engine: "Opus",
+            engine: "Bergamot",
             from: "en",
-            to: "it",
+            to: "es",
         },
         onProgress: (p) => console.log(`   Loading NMT: ${p.percentage.toFixed(1)}%`),
     });

package/dist/examples/suspend-resume.js

@@ -1,4 +1,4 @@
-import { loadModel, LLAMA_3_2_1B_INST_Q4_0, completion, unloadModel, suspend, resume, } from "@qvac/sdk";
+import { loadModel, LLAMA_3_2_1B_INST_Q4_0, completion, unloadModel, suspend, resume, state, } from "@qvac/sdk";
 try {
     // Load a model
     const modelId = await loadModel({
@@ -9,6 +9,7 @@ try {
         },
     });
     console.log("✅ Model loaded\n");
+    console.log(`📊 Lifecycle state: ${await state()}\n`);
     // Run a completion before suspending
     console.log("--- Completion before suspend ---");
     const result1 = completion({
@@ -23,14 +24,30 @@ try {
     // Suspend all networking and storage (e.g. app going to background)
     console.log("⏸️  Suspending...");
     await suspend();
-    console.log("✅ Suspended — Hyperswarm and Corestore resources are paused\n");
+    console.log(`📊 Lifecycle state: ${await state()}\n`);
+    try {
+        await completion({
+            modelId,
+            history: [{ role: "user", content: "This should fail" }],
+            stream: false,
+        }).text;
+    }
+    catch (error) {
+        const name = error.name;
+        if (name === "LIFECYCLE_OPERATION_BLOCKED") {
+            console.log(`🚫 Operation blocked while suspended (${name})`);
+        }
+        else {
+            throw error;
+        }
+    }
     // Simulate time in background
-    console.log("💤 Simulating 3 seconds in background...");
+    console.log("\n💤 Simulating 3 seconds in background...");
     await new Promise((resolve) => setTimeout(resolve, 3000));
     // Resume when returning to foreground
     console.log("▶️  Resuming...");
     await resume();
-    console.log("✅ Resumed — networking and storage restored\n");
+    console.log(`📊 Lifecycle state: ${await state()}\n`);
     // Run another completion after resuming
     console.log("--- Completion after resume ---");
     const result2 = completion({

package/dist/examples/transcription/parakeet-microphone-record.js

@@ -9,29 +9,12 @@
  * Requirements: FFmpeg installed, microphone access.
  */
 import { loadModel, unloadModel, transcribe, PARAKEET_TDT_ENCODER_FP32, PARAKEET_TDT_DECODER_FP32, PARAKEET_TDT_VOCAB, PARAKEET_TDT_PREPROCESSOR_FP32, } from "@qvac/sdk";
-import { spawn, spawnSync } from "child_process";
-import { platform } from "os";
+import { spawnSync } from "child_process";
+import { startMicrophone } from "../audio/mic-input";
 const SAMPLE_RATE = 16000;
 const BYTES_PER_SAMPLE = 2; // s16le
 const CHUNK_DURATION_S = 3;
 const CHUNK_SIZE = SAMPLE_RATE * BYTES_PER_SAMPLE * CHUNK_DURATION_S;
-function getAudioInputArgs() {
-    switch (platform()) {
-        case "darwin":
-            return ["-f", "avfoundation", "-i", ":0"];
-        case "win32":
-            return [
-                "-f",
-                "dshow",
-                "-i",
-                "audio=@device_cm_{33D9A762-90C8-11D0-BD43-00A0C911CE86}\\wave_{58C07110-A4FD-4FF8-BA10-5A3C14389F71}",
-            ];
-        case "linux":
-            return ["-f", "pulse", "-i", "default"];
-        default:
-            throw new Error(`Unsupported platform: ${platform()}`);
-    }
-}
 // ── Main ──
 try {
     const r = spawnSync("ffmpeg", ["-version"], { stdio: "ignore" });
@@ -55,20 +38,10 @@ const modelId = await loadModel({
     onProgress: (p) => console.log(`Download: ${p.percentage.toFixed(1)}%`),
 });
 console.log("Model loaded.\n");
-const ffmpeg = spawn("ffmpeg", [
-    ...getAudioInputArgs(),
-    "-ar",
-    String(SAMPLE_RATE),
-    "-ac",
-    "1",
-    "-sample_fmt",
-    "s16",
-    "-f",
-    "s16le",
-    "pipe:1",
-], { stdio: ["ignore", "pipe", "ignore"] });
-if (!ffmpeg.stdout)
-    throw new Error("Failed to open microphone");
+const ffmpeg = startMicrophone({
+    sampleRate: SAMPLE_RATE,
+    format: "s16le",
+});
 let buffer = Buffer.alloc(0);
 let processing = false;
 console.log("Listening... speak and pause to see transcriptions.\n");

package/dist/examples/transcription/whispercpp-filesystem-streaming.js

@@ -51,7 +51,7 @@ try {
     });
     console.log(`Model loaded: ${modelId}\n`);
     console.log("Opening live session...");
-    const session = await transcribeStream({ modelId });
+    const session = await transcribeStream({ modelId, metadata: true });
     console.log("Session open. Streaming audio...\n");
     const ffmpeg = spawn("ffmpeg", [
         "-i", SAMPLE_FILE,
@@ -76,14 +76,16 @@ try {
         session.end();
     });
     const segments = [];
-    for await (const text of session) {
-        segments.push(text.trim());
-        console.log(`  [${segments.length}] ${text.trim()}`);
+    for await (const segment of session) {
+        segments.push(segment);
+        const start = (segment.startMs / 1000).toFixed(2);
+        const end = (segment.endMs / 1000).toFixed(2);
+        console.log(`  [${segments.length}] [${start}s → ${end}s] (id=${segment.id}, append=${segment.append}) ${segment.text.trim()}`);
     }
     console.log("\n=== Results ===");
     console.log(`Segments: ${segments.length}`);
     if (segments.length > 0) {
-        console.log(`Transcript: ${segments.join(" ")}`);
+        console.log(`Transcript: ${segments.map((s) => s.text.trim()).join(" ")}`);
     }
     else {
         console.log("WARNING: No transcription segments received!");

package/dist/examples/transcription/whispercpp-filesystem.js

@@ -53,11 +53,20 @@ try {
         },
     });
     console.log(`✅ Whisper model loaded with ID: ${modelId}`);
-    // Perform transcription
+    // Perform transcription with per-segment metadata
     console.log("🎧 Transcribing audio...");
-    const text = await transcribe({ modelId, audioChunk: audioFilePath });
+    const segments = await transcribe({
+        modelId,
+        audioChunk: audioFilePath,
+        metadata: true,
+    });
     console.log("📝 Transcription result:");
-    console.log(text);
+    for (const segment of segments) {
+        const start = (segment.startMs / 1000).toFixed(2);
+        const end = (segment.endMs / 1000).toFixed(2);
+        console.log(`  [${start}s → ${end}s] (id=${segment.id}, append=${segment.append}) ${segment.text}`);
+    }
+    console.log(`\nFull transcript: ${segments.map((s) => s.text).join("").trim()}`);
     // Unload the model when done
     console.log("🧹 Unloading Whisper model...");
     await unloadModel({ modelId });

package/dist/examples/transcription/whispercpp-microphone-record.js

@@ -9,26 +9,9 @@
  * Requirements: FFmpeg installed, microphone access.
  */
 import { loadModel, unloadModel, transcribeStream, WHISPER_TINY, VAD_SILERO_5_1_2, } from "@qvac/sdk";
-import { spawn, spawnSync } from "child_process";
-import { platform } from "os";
+import { spawnSync } from "child_process";
+import { startMicrophone } from "../audio/mic-input";
 const SAMPLE_RATE = 16000;
-function getAudioInputArgs() {
-    switch (platform()) {
-        case "darwin":
-            return ["-f", "avfoundation", "-i", ":0"];
-        case "win32":
-            return [
-                "-f",
-                "dshow",
-                "-i",
-                "audio=@device_cm_{33D9A762-90C8-11D0-BD43-00A0C911CE86}\\wave_{58C07110-A4FD-4FF8-BA10-5A3C14389F71}",
-            ];
-        case "linux":
-            return ["-f", "pulse", "-i", "default"];
-        default:
-            throw new Error(`Unsupported platform: ${platform()}`);
-    }
-}
 // ── Main ──
 try {
     const r = spawnSync("ffmpeg", ["-version"], { stdio: "ignore" });
@@ -63,16 +46,10 @@ const modelId = await loadModel({
     },
 });
 console.log("Model loaded.\n");
-const ffmpeg = spawn("ffmpeg", [
-    ...getAudioInputArgs(),
-    "-ar", String(SAMPLE_RATE),
-    "-ac", "1",
-    "-sample_fmt", "flt",
-    "-f", "f32le",
-    "pipe:1",
-], { stdio: ["ignore", "pipe", "ignore"] });
-if (!ffmpeg.stdout)
-    throw new Error("Failed to open microphone");
+const ffmpeg = startMicrophone({
+    sampleRate: SAMPLE_RATE,
+    format: "f32le",
+});
 const session = await transcribeStream({ modelId });
 ffmpeg.stdout.on("data", (chunk) => session.write(chunk));
 console.log("Listening... speak and pause to see transcriptions.\n");

package/dist/examples/translation/translation-stream.js

@@ -1,12 +1,12 @@
-import { loadModel, translate, unloadModel, MARIAN_OPUS_EN_IT_Q4_0, } from "@qvac/sdk";
+import { loadModel, translate, unloadModel, BERGAMOT_EN_ES, } from "@qvac/sdk";
 try {
     const modelId = await loadModel({
-        modelSrc: MARIAN_OPUS_EN_IT_Q4_0,
+        modelSrc: BERGAMOT_EN_ES,
         modelType: "nmt",
         modelConfig: {
-            engine: "Opus",
+            engine: "Bergamot",
             from: "en",
-            to: "it",
+            to: "es",
         },
     });
     console.log(`✅ Model loaded: ${modelId}`);
@@ -18,7 +18,7 @@ try {
         modelType: "nmt",
         stream: true,
     });
-    process.stdout.write("Translated text EN -> IT: ");
+    process.stdout.write("Translated text EN -> ES: ");
     for await (const token of streamResult.tokenStream) {
         process.stdout.write(token);
     }

package/dist/examples/tts/llm-to-tts-streaming.d.ts

@@ -0,0 +1,16 @@
+/**
+ * Stream an LLM and forward token deltas through the duplex TTS path (`textToSpeechStream` →
+ * `runStreaming` with `accumulateSentences: true`). Text is filtered before TTS / console: any
+ * `<...>` span (including across tokens) is removed, and `*` characters are stripped.
+ *
+ * Each synthesized phrase is played as soon as PCM arrives (`playPcmInt16Chunk`). A combined WAV
+ * is written at the end for inspection.
+ *
+ * Prerequisites: Bun, QVAC worker, registry access, macOS `afplay` / Linux `aplay` (or Windows
+ * PowerShell player) for chunk playback.
+ *
+ * Usage:
+ *   bun run examples/tts/llm-to-tts-streaming.ts
+ */
+export {};
+//# sourceMappingURL=llm-to-tts-streaming.d.ts.map

package/dist/examples/tts/llm-to-tts-streaming.js

@@ -0,0 +1,143 @@
+/**
+ * Stream an LLM and forward token deltas through the duplex TTS path (`textToSpeechStream` →
+ * `runStreaming` with `accumulateSentences: true`). Text is filtered before TTS / console: any
+ * `<...>` span (including across tokens) is removed, and `*` characters are stripped.
+ *
+ * Each synthesized phrase is played as soon as PCM arrives (`playPcmInt16Chunk`). A combined WAV
+ * is written at the end for inspection.
+ *
+ * Prerequisites: Bun, QVAC worker, registry access, macOS `afplay` / Linux `aplay` (or Windows
+ * PowerShell player) for chunk playback.
+ *
+ * Usage:
+ *   bun run examples/tts/llm-to-tts-streaming.ts
+ */
+import { completion, loadModel, textToSpeechStream, unloadModel, LLAMA_3_2_1B_INST_Q4_0, TTS_SUPERTONIC2_OFFICIAL_TEXT_ENCODER_SUPERTONE_FP32, TTS_SUPERTONIC2_OFFICIAL_DURATION_PREDICTOR_SUPERTONE_FP32, TTS_SUPERTONIC2_OFFICIAL_VECTOR_ESTIMATOR_SUPERTONE_FP32, TTS_SUPERTONIC2_OFFICIAL_VOCODER_SUPERTONE_FP32, TTS_SUPERTONIC2_OFFICIAL_UNICODE_INDEXER_SUPERTONE_FP32, TTS_SUPERTONIC2_OFFICIAL_TTS_CONFIG_SUPERTONE, TTS_SUPERTONIC2_OFFICIAL_VOICE_STYLE_SUPERTONE, } from "@qvac/sdk";
+import { createWav, playPcmInt16Chunk } from "./utils.js";
+const SUPERTONIC_SAMPLE_RATE = 44100;
+/**
+ * Drop angle-bracket spans (e.g. `<think>…</think>`, `<|foo|>`) even when split across tokens,
+ * and strip `*` (markdown emphasis) from text passed to TTS and the console.
+ */
+function createAngleBracketAndStarFilter() {
+    let inAngle = false;
+    return function filterChunk(chunk) {
+        let out = "";
+        for (let i = 0; i < chunk.length; i++) {
+            const c = chunk[i];
+            if (inAngle) {
+                if (c === ">") {
+                    inAngle = false;
+                }
+                continue;
+            }
+            if (c === "<") {
+                inAngle = true;
+                continue;
+            }
+            if (c === "*") {
+                continue;
+            }
+            out += c;
+        }
+        return out;
+    };
+}
+/**
+ * Append PCM without `push(...huge)` — a spread of hundreds of thousands of
+ * arguments blows the V8 / Bare argument-count limit. Batching with
+ * `Array.prototype.push.apply` is both faster than `push(...slice)` and
+ * avoids allocating an intermediate rest array on the stack per call.
+ */
+function appendPcmSamples(target, chunk) {
+    const batch = 8192;
+    for (let i = 0; i < chunk.length; i += batch) {
+        const end = Math.min(i + batch, chunk.length);
+        Array.prototype.push.apply(target, chunk.slice(i, end));
+    }
+}
+try {
+    console.log(`Loading LLM from registry: ${LLAMA_3_2_1B_INST_Q4_0.name}`);
+    const llmModelId = await loadModel({
+        modelSrc: LLAMA_3_2_1B_INST_Q4_0.src,
+        modelType: "llm",
+        modelConfig: {
+            ctx_size: 2048,
+        },
+        onProgress: (progress) => console.log(`LLM load: ${progress.percentage.toFixed(1)}%`),
+    });
+    console.log(`LLM ready: ${llmModelId}`);
+    console.log("Loading Supertonic TTS (registry)…");
+    const ttsModelId = await loadModel({
+        modelSrc: TTS_SUPERTONIC2_OFFICIAL_TEXT_ENCODER_SUPERTONE_FP32.src,
+        modelType: "tts",
+        modelConfig: {
+            ttsEngine: "supertonic",
+            language: "en",
+            ttsSpeed: 1.05,
+            ttsNumInferenceSteps: 10,
+            ttsSupertonicMultilingual: true,
+            ttsTextEncoderSrc: TTS_SUPERTONIC2_OFFICIAL_TEXT_ENCODER_SUPERTONE_FP32.src,
+            ttsDurationPredictorSrc: TTS_SUPERTONIC2_OFFICIAL_DURATION_PREDICTOR_SUPERTONE_FP32.src,
+            ttsVectorEstimatorSrc: TTS_SUPERTONIC2_OFFICIAL_VECTOR_ESTIMATOR_SUPERTONE_FP32.src,
+            ttsVocoderSrc: TTS_SUPERTONIC2_OFFICIAL_VOCODER_SUPERTONE_FP32.src,
+            ttsUnicodeIndexerSrc: TTS_SUPERTONIC2_OFFICIAL_UNICODE_INDEXER_SUPERTONE_FP32.src,
+            ttsTtsConfigSrc: TTS_SUPERTONIC2_OFFICIAL_TTS_CONFIG_SUPERTONE.src,
+            ttsVoiceStyleSrc: TTS_SUPERTONIC2_OFFICIAL_VOICE_STYLE_SUPERTONE.src,
+        },
+        onProgress: (progress) => console.log(`TTS load: ${progress.percentage.toFixed(1)}%`),
+    });
+    console.log(`TTS ready: ${ttsModelId}`);
+    const prompt = "What is a constellation?";
+    console.log(`\nUser: ${prompt}\nAssistant (streaming):`);
+    const result = completion({
+        modelId: llmModelId,
+        history: [{ role: "user", content: prompt }],
+        stream: true,
+    });
+    const combinedPcm = [];
+    const ttsSession = await textToSpeechStream({
+        modelId: ttsModelId,
+        inputType: "text",
+        accumulateSentences: true,
+    });
+    let phraseIndex = 0;
+    const filterChunk = createAngleBracketAndStarFilter();
+    const drainPcm = (async () => {
+        for await (const m of ttsSession) {
+            if (m.buffer.length > 0) {
+                appendPcmSamples(combinedPcm, m.buffer);
+                phraseIndex += 1;
+                const preview = typeof m.sentenceChunk === "string"
+                    ? m.sentenceChunk.replace(/\s+/g, " ").trim().slice(0, 72)
+                    : "";
+                console.log(`\n[TTS phrase ${phraseIndex}] ${m.buffer.length} samples${preview ? ` — "${preview}${preview.length >= 72 ? "..." : ""}"` : ""}`);
+                await playPcmInt16Chunk(m.buffer, SUPERTONIC_SAMPLE_RATE);
+            }
+        }
+    })();
+    for await (const token of result.tokenStream) {
+        const cleaned = filterChunk(token);
+        if (cleaned.length > 0) {
+            process.stdout.write(cleaned);
+            ttsSession.write(cleaned);
+        }
+    }
+    ttsSession.end();
+    await drainPcm;
+    const stats = await result.stats;
+    if (stats) {
+        console.log("\nLLM stats:", stats);
+    }
+    const outWav = "llm-to-tts-streaming-output.wav";
+    console.log(`\nWriting ${combinedPcm.length} samples to ${outWav} (full utterance; phrases were already played above).`);
+    createWav(combinedPcm, SUPERTONIC_SAMPLE_RATE, outWav);
+    await unloadModel({ modelId: llmModelId, clearStorage: false });
+    await unloadModel({ modelId: ttsModelId, clearStorage: false });
+    console.log("Models unloaded.");
+    process.exit(0);
+}
+catch (error) {
+    console.error("❌ Error:", error);
+    process.exit(1);
+}

package/dist/examples/tts/supertonic-multilingual.js

@@ -10,9 +10,9 @@ try {
         modelConfig: {
             ttsEngine: "supertonic",
             language: "es",
-            speed: 1.05,
-            numInferenceSteps: 5,
-            supertonicMultilingual: true, //false for English quality.
+            ttsSpeed: 1.05,
+            ttsNumInferenceSteps: 5,
+            ttsSupertonicMultilingual: true, //false for English quality.
             ttsTextEncoderSrc: TTS_SUPERTONIC2_OFFICIAL_TEXT_ENCODER_SUPERTONE_FP32.src,
             ttsDurationPredictorSrc: TTS_SUPERTONIC2_OFFICIAL_DURATION_PREDICTOR_SUPERTONE_FP32.src,
             ttsVectorEstimatorSrc: TTS_SUPERTONIC2_OFFICIAL_VECTOR_ESTIMATOR_SUPERTONE_FP32.src,

package/dist/examples/tts/supertonic.js

@@ -10,9 +10,9 @@ try {
         modelConfig: {
             ttsEngine: "supertonic",
             language: "en",
-            speed: 1.05,
-            numInferenceSteps: 5,
-            supertonicMultilingual: false, //false for English quality.
+            ttsSpeed: 1.05,
+            ttsNumInferenceSteps: 5,
+            ttsSupertonicMultilingual: false, //false for English quality.
             ttsTextEncoderSrc: TTS_SUPERTONIC2_OFFICIAL_TEXT_ENCODER_SUPERTONE_FP32.src,
             ttsDurationPredictorSrc: TTS_SUPERTONIC2_OFFICIAL_DURATION_PREDICTOR_SUPERTONE_FP32.src,
             ttsVectorEstimatorSrc: TTS_SUPERTONIC2_OFFICIAL_VECTOR_ESTIMATOR_SUPERTONE_FP32.src,