handy-remote-server 1.2.0 → 1.3.0

package/README.md

@@ -1,6 +1,6 @@
 # Handy Remote Server 🎙️
 
-A lightweight standalone inference server for [Handy](https://github.com/cjpais/Handy), allowing you to transcribe audio from external devices, weak computers, and more.
+A lightweight standalone inference server for [Handy](https://github.com/viktor-silakov/Handy), allowing you to transcribe audio from external devices, weak computers, and more.
 
 ## Installation
 
@@ -67,11 +67,47 @@ Handy Remote Server is running on port 3000
 
 ## Environment Variables
 
-| Variable         | Default                                     | Description                     |
-| ---------------- | ------------------------------------------- | ------------------------------- |
-| `PORT`           | `3000`                                      | Server port                     |
-| `API_KEY`        | auto-generated, saved to `~/.handy/api_key` | Bearer token for authentication |
-| `INFER_CLI_PATH` | auto-detected                               | Path to the `rust-infer` binary |
+| Variable         | Default                                     | Description                            |
+| ---------------- | ------------------------------------------- | -------------------------------------- |
+| `PORT`           | `3000`                                      | Server port                            |
+| `API_KEY`        | auto-generated, saved to `~/.handy/api_key` | Bearer token for authentication        |
+| `INFER_CLI_PATH` | auto-detected                               | Path to the `rust-infer` binary        |
+| `MODEL_TYPE`     | `gigaam`                                    | Transcription model to use (see below) |
+
+## Supported Models
+
+```bash
+# Русский (по умолчанию)
+MODEL_TYPE=gigaam npx handy-remote-server
+
+# Мультиязычный (включая русский) — Whisper модели
+MODEL_TYPE=whisper-tiny npx handy-remote-server      # 75 MB
+MODEL_TYPE=whisper-base npx handy-remote-server      # 142 MB
+MODEL_TYPE=whisper-small npx handy-remote-server     # 487 MB
+MODEL_TYPE=whisper-medium npx handy-remote-server    # 1.5 GB
+
+# Английский — Moonshine
+MODEL_TYPE=moonshine-tiny npx handy-remote-server    # 60 MB
+MODEL_TYPE=moonshine-base npx handy-remote-server    # 100 MB
+
+# Английский — Breeze/Parakeet
+MODEL_TYPE=parakeet npx handy-remote-server          # ~200 MB
+
+# Мультиязычный — SenseVoice
+MODEL_TYPE=sensevoice npx handy-remote-server        # ~200 MB
+```
+
+| Model              | Language       | Size    | Speed     |
+| ------------------ | -------------- | ------- | --------- |
+| `gigaam` (default) | Russian        | ~100 MB | ⚡ Fast   |
+| `whisper-tiny`     | Multi-language | 75 MB   | ⚡ Fast   |
+| `whisper-base`     | Multi-language | 142 MB  | ⚡ Fast   |
+| `whisper-small`    | Multi-language | 487 MB  | 🔄 Medium |
+| `whisper-medium`   | Multi-language | 1.5 GB  | 🐢 Slow   |
+| `moonshine-tiny`   | English        | 60 MB   | ⚡ Fast   |
+| `moonshine-base`   | English        | 100 MB  | ⚡ Fast   |
+| `parakeet`         | English        | ~200 MB | 🔄 Medium |
+| `sensevoice`       | Multi-language | ~200 MB | 🔄 Medium |
 
 ## How It Works
 

package/dist/index.js

@@ -5,13 +5,14 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
 };
 Object.defineProperty(exports, "__esModule", { value: true });
 const express_1 = __importDefault(require("express"));
-const multer_1 = __importDefault(require("multer"));
 const child_process_1 = require("child_process");
 const crypto_1 = __importDefault(require("crypto"));
 const path_1 = __importDefault(require("path"));
 const fs_1 = __importDefault(require("fs"));
 const os_1 = __importDefault(require("os"));
 const dotenv_1 = __importDefault(require("dotenv"));
+const tar_fs_1 = __importDefault(require("tar-fs"));
+const gunzip_maybe_1 = __importDefault(require("gunzip-maybe"));
 dotenv_1.default.config();
 const app = (0, express_1.default)();
 const port = process.env.PORT || 3000;
@@ -19,36 +20,21 @@ const port = process.env.PORT || 3000;
 const handyDir = path_1.default.join(os_1.default.homedir(), '.handy');
 const keyFilePath = path_1.default.join(handyDir, 'api_key');
 function loadOrCreateApiKey() {
-    // 1. Env var takes priority
-    if (process.env.API_KEY) {
+    if (process.env.API_KEY)
         return process.env.API_KEY;
-    }
-    // 2. Try to load from cached file
     if (fs_1.default.existsSync(keyFilePath)) {
         const cached = fs_1.default.readFileSync(keyFilePath, 'utf-8').trim();
-        if (cached.length > 0) {
-            console.log(`\n======================================================`);
-            console.log(`Loaded API KEY from ${keyFilePath}`);
-            console.log(`Your API KEY is: ${cached}`);
-            console.log(`======================================================\n`);
+        if (cached.length > 0)
             return cached;
-        }
     }
-    // 3. Generate a new one and persist it
     const newKey = crypto_1.default.randomBytes(32).toString('hex');
     fs_1.default.mkdirSync(handyDir, { recursive: true });
     fs_1.default.writeFileSync(keyFilePath, newKey + '\n', { mode: 0o600 });
-    console.log(`\n======================================================`);
-    console.log(`Generated a new API KEY (saved to ${keyFilePath})`);
-    console.log(`Your API KEY is: ${newKey}`);
-    console.log(`======================================================\n`);
     return newKey;
 }
 const API_KEY = loadOrCreateApiKey();
 // ── Logging helpers ───────────────────────────────────────────────────
-function timestamp() {
-    return new Date().toISOString();
-}
+function timestamp() { return new Date().toISOString(); }
 function formatBytes(bytes) {
     if (bytes < 1024)
         return `${bytes} B`;
@@ -61,94 +47,98 @@ function formatDuration(ms) {
         return `${ms}ms`;
     return `${(ms / 1000).toFixed(2)}s`;
 }
-// ── Ensure directories ────────────────────────────────────────────────
-const modelsDir = path_1.default.join(__dirname, '..', 'models');
-if (!fs_1.default.existsSync(modelsDir)) {
-    fs_1.default.mkdirSync(modelsDir, { recursive: true });
+const MODEL_REGISTRY = {
+    'gigaam': {
+        engine: 'gigaam',
+        url: 'https://blob.handy.computer/giga-am-v3.int8.onnx',
+        filename: 'gigaam.onnx'
+    },
+    'whisper-tiny': {
+        engine: 'whisper',
+        url: 'https://blob.handy.computer/ggml-tiny.bin',
+        filename: 'whisper-tiny.bin'
+    },
+    'whisper-base': {
+        engine: 'whisper',
+        url: 'https://blob.handy.computer/ggml-base.bin',
+        filename: 'whisper-base.bin'
+    },
+    'whisper-small': {
+        engine: 'whisper',
+        url: 'https://blob.handy.computer/ggml-small.bin',
+        filename: 'whisper-small.bin'
+    },
+    'whisper-medium': {
+        engine: 'whisper',
+        url: 'https://blob.handy.computer/whisper-medium-q4_1.bin',
+        filename: 'whisper-medium.bin'
+    },
+    'moonshine-tiny': {
+        engine: 'moonshine',
+        url: 'https://blob.handy.computer/moonshine-tiny-streaming-en.tar.gz',
+        filename: 'moonshine-tiny', // Dir name after extraction
+        isArchive: true
+    },
+    'moonshine-base': {
+        engine: 'moonshine',
+        url: 'https://blob.handy.computer/moonshine-base.tar.gz',
+        filename: 'moonshine-base',
+        isArchive: true
+    },
+    'parakeet': {
+        engine: 'parakeet',
+        url: 'https://blob.handy.computer/parakeet-v3-int8.tar.gz',
+        filename: 'parakeet-v3',
+        isArchive: true,
+        configFilename: 'preprocessor.json'
+    },
+    'sensevoice': {
+        engine: 'sensevoice',
+        url: 'https://blob.handy.computer/sense-voice-int8.tar.gz',
+        filename: 'sensevoice',
+        isArchive: true
+    }
+};
+const SELECTED_MODEL_TYPE = (process.env.MODEL_TYPE || 'gigaam').toLowerCase();
+const modelCfg = MODEL_REGISTRY[SELECTED_MODEL_TYPE];
+if (!modelCfg) {
+    console.error(`Error: Unknown MODEL_TYPE "${SELECTED_MODEL_TYPE}".`);
+    console.error(`Supported types: ${Object.keys(MODEL_REGISTRY).join(', ')}`);
+    process.exit(1);
 }
+// ── Directories ───────────────────────────────────────────────────────
+const modelsBaseDir = path_1.default.join(__dirname, '..', 'models');
 const uploadDir = path_1.default.join(__dirname, '..', 'uploads');
-if (!fs_1.default.existsSync(uploadDir)) {
-    fs_1.default.mkdirSync(uploadDir, { recursive: true });
-}
-// ── Request logging middleware ────────────────────────────────────────
-let requestCounter = 0;
-app.use((req, res, next) => {
-    const reqId = ++requestCounter;
-    const start = Date.now();
-    const ip = req.headers['x-forwarded-for'] || req.socket.remoteAddress || 'unknown';
-    console.log(`\n[${timestamp()}] ── REQUEST #${reqId} ──────────────────────`);
-    console.log(`  Method:  ${req.method} ${req.path}`);
-    console.log(`  From:    ${ip}`);
-    console.log(`  Headers: Content-Type=${req.headers['content-type'] || 'N/A'}, Content-Length=${req.headers['content-length'] || 'N/A'}`);
-    // Store metadata on request for later use
-    req._reqId = reqId;
-    req._startTime = start;
-    req._ip = ip;
-    const originalJson = res.json.bind(res);
-    res.json = function (body) {
-        const duration = Date.now() - start;
-        const status = res.statusCode;
-        console.log(`[${timestamp()}] ── RESPONSE #${reqId} ─────────────────────`);
-        console.log(`  Status:   ${status}`);
-        console.log(`  Duration: ${formatDuration(duration)}`);
-        if (body?.text) {
-            const preview = body.text.length > 100 ? body.text.substring(0, 100) + '...' : body.text;
-            console.log(`  Result:   "${preview}"`);
-        }
-        else if (body?.error) {
-            console.log(`  Error:    ${body.error}`);
-        }
-        console.log(`  ────────────────────────────────────────────────`);
-        return originalJson(body);
-    };
-    next();
-});
-// ── Authentication middleware ─────────────────────────────────────────
-app.use((req, res, next) => {
-    const authHeader = req.headers.authorization;
-    if (!authHeader || !authHeader.startsWith('Bearer ')) {
-        console.log(`  Auth:     REJECTED (missing/invalid Authorization header)`);
-        return res.status(401).json({ error: 'Missing or invalid Authorization header' });
-    }
-    const token = authHeader.split(' ')[1];
-    if (token !== API_KEY) {
-        console.log(`  Auth:     REJECTED (invalid key)`);
-        return res.status(403).json({ error: 'Invalid API Key' });
-    }
-    console.log(`  Auth:     OK`);
-    next();
-});
-// ── Multer storage ────────────────────────────────────────────────────
-const storage = multer_1.default.diskStorage({
-    destination: function (req, file, cb) {
-        cb(null, uploadDir);
-    },
-    filename: function (req, file, cb) {
-        const uniqueSuffix = Date.now() + '-' + Math.round(Math.random() * 1E9);
-        cb(null, file.fieldname + '-' + uniqueSuffix + '.wav');
+[modelsBaseDir, uploadDir].forEach(d => { if (!fs_1.default.existsSync(d))
+    fs_1.default.mkdirSync(d, { recursive: true }); });
+// ── Model paths ───────────────────────────────────────────────────────
+const modelPath = path_1.default.join(modelsBaseDir, modelCfg.filename);
+let actualModelFile = modelPath;
+let parakeetConfigPath = '';
+if (modelCfg.isArchive) {
+    // For archives, we look for model.onnx inside the directory
+    actualModelFile = path_1.default.join(modelPath, 'model.onnx');
+    if (modelCfg.engine === 'parakeet') {
+        parakeetConfigPath = path_1.default.join(modelPath, modelCfg.configFilename);
     }
-});
-const upload = (0, multer_1.default)({ storage: storage });
-// ── Model download ───────────────────────────────────────────────────
-const GIGAAM_MODEL_URL = 'https://blob.handy.computer/giga-am-v3.int8.onnx';
-const gigaamModelPath = path_1.default.join(modelsDir, 'gigaam.onnx');
-async function downloadFile(url, dest) {
-    if (fs_1.default.existsSync(dest))
+}
+// ── Download & Extract ────────────────────────────────────────────────
+async function downloadAndPrepare() {
+    if (fs_1.default.existsSync(actualModelFile))
         return;
-    console.log(`\n📥 Downloading model...`);
-    console.log(`   URL:  ${url}`);
-    console.log(`   Dest: ${dest}\n`);
-    fs_1.default.mkdirSync(path_1.default.dirname(dest), { recursive: true });
-    const response = await fetch(url);
+    const dest = modelCfg.isArchive ? modelPath + '.tar.gz' : modelPath;
+    console.log(`\n📥 Downloading model: ${SELECTED_MODEL_TYPE}...`);
+    console.log(`   URL:  ${modelCfg.url}`);
+    const response = await fetch(modelCfg.url);
     if (!response.ok)
-        throw new Error(`Failed to fetch ${url}: ${response.statusText}`);
+        throw new Error(`Failed to fetch: ${response.statusText}`);
     const totalBytes = parseInt(response.headers.get('content-length') || '0', 10);
     let downloadedBytes = 0;
     const startTime = Date.now();
     const fileStream = fs_1.default.createWriteStream(dest);
     const reader = response.body?.getReader();
     if (!reader)
-        throw new Error('Response body is not readable');
+        throw new Error('Body not readable');
     const barWidth = 40;
     while (true) {
         const { done, value } = await reader.read();
@@ -156,69 +146,92 @@ async function downloadFile(url, dest) {
             break;
         fileStream.write(Buffer.from(value));
         downloadedBytes += value.length;
-        // Draw progress bar
         const percent = totalBytes > 0 ? downloadedBytes / totalBytes : 0;
         const filled = Math.round(barWidth * percent);
-        const empty = barWidth - filled;
-        const bar = '█'.repeat(filled) + '░'.repeat(empty);
+        const bar = '█'.repeat(filled) + '░'.repeat(barWidth - filled);
         const pct = (percent * 100).toFixed(1).padStart(5);
-        const dl = formatBytes(downloadedBytes);
-        const tot = totalBytes > 0 ? formatBytes(totalBytes) : '?';
-        const elapsed = (Date.now() - startTime) / 1000;
-        const speed = elapsed > 0 ? formatBytes(downloadedBytes / elapsed) + '/s' : '';
-        process.stdout.write(`\r   ${bar} ${pct}%  ${dl} / ${tot}  ${speed}   `);
+        const speed = (downloadedBytes / ((Date.now() - startTime) / 1000) / 1024 / 1024).toFixed(1);
+        process.stdout.write(`\r   ${bar} ${pct}%  ${formatBytes(downloadedBytes)} / ${formatBytes(totalBytes)}  ${speed} MB/s   `);
     }
-    await new Promise((resolve, reject) => {
-        fileStream.end(() => resolve());
-        fileStream.on('error', reject);
-    });
-    const totalTime = ((Date.now() - startTime) / 1000).toFixed(1);
+    await new Promise(r => fileStream.end(() => r()));
     process.stdout.write('\n');
-    console.log(`\n✅ Download complete in ${totalTime}s\n`);
-}
-async function ensureModels() {
-    await downloadFile(GIGAAM_MODEL_URL, gigaamModelPath);
+    if (modelCfg.isArchive) {
+        console.log(`📦 Extracting archive to ${modelPath}...`);
+        fs_1.default.mkdirSync(modelPath, { recursive: true });
+        await new Promise((resolve, reject) => {
+            fs_1.default.createReadStream(dest)
+                .pipe((0, gunzip_maybe_1.default)())
+                .pipe(tar_fs_1.default.extract(modelPath))
+                .on('finish', resolve)
+                .on('error', reject);
+        });
+        fs_1.default.unlinkSync(dest); // Cleanup
+    }
+    console.log(`✅ Ready!\n`);
 }
+// ── Request logging ───────────────────────────────────────────────────
+let requestCounter = 0;
+app.use((req, res, next) => {
+    const reqId = ++requestCounter;
+    const start = Date.now();
+    const ip = req.headers['x-forwarded-for'] || req.socket.remoteAddress || 'unknown';
+    console.log(`\n[${timestamp()}] ── REQUEST #${reqId} ──────────────────────`);
+    console.log(`  Method:  ${req.method} ${req.path}`);
+    console.log(`  Model:   ${SELECTED_MODEL_TYPE}`);
+    req._reqId = reqId;
+    req._startTime = start;
+    const originalJson = res.json.bind(res);
+    res.json = function (body) {
+        console.log(`[${timestamp()}] ── RESPONSE #${reqId} (Status: ${res.statusCode}, ${Date.now() - start}ms) ─────`);
+        if (body?.text)
+            console.log(`  Result:   "${body.text.substring(0, 100)}${body.text.length > 100 ? '...' : ''}"`);
+        else if (body?.error)
+            console.log(`  Error:    ${body.error}`);
+        return originalJson(body);
+    };
+    next();
+});
+// ── Auth ──────────────────────────────────────────────────────────────
+app.use((req, res, next) => {
+    const authHeader = req.headers.authorization;
+    if (!authHeader?.startsWith('Bearer ') || authHeader.split(' ')[1] !== API_KEY) {
+        return res.status(401).json({ error: 'Auth failed' });
+    }
+    next();
+});
+// ── Inference Bridge ──────────────────────────────────────────────────
 let inferProcess = null;
 let isReady = false;
 let resolvers = {};
-ensureModels().then(() => {
-    let inferProcessPath = process.env.INFER_CLI_PATH || path_1.default.join(__dirname, '..', 'rust-infer', 'target', 'release', 'rust-infer');
-    if (!fs_1.default.existsSync(inferProcessPath)) {
-        inferProcessPath = path_1.default.join(__dirname, '..', 'rust-infer', 'target', 'debug', 'rust-infer');
-    }
-    console.log(`Using inference CLI: ${inferProcessPath}`);
-    inferProcess = (0, child_process_1.spawn)(inferProcessPath, [gigaamModelPath], { stdio: ['pipe', 'pipe', 'inherit'] });
+downloadAndPrepare().then(() => {
+    let binPath = process.env.INFER_CLI_PATH || path_1.default.join(__dirname, '..', 'rust-infer', 'target', 'release', 'rust-infer');
+    if (!fs_1.default.existsSync(binPath))
+        binPath = path_1.default.join(__dirname, '..', 'rust-infer', 'target', 'debug', 'rust-infer');
+    console.log(`Starting inference: ${binPath}`);
+    const args = [modelCfg.engine, actualModelFile];
+    if (parakeetConfigPath)
+        args.push(parakeetConfigPath);
+    inferProcess = (0, child_process_1.spawn)(binPath, args, { stdio: ['pipe', 'pipe', 'inherit'] });
     inferProcess.stdout.on('data', (data) => {
-        const lines = data.toString().split('\n').map((l) => l.trim()).filter(Boolean);
-        for (const line of lines) {
-            if (line === 'READY') {
+        data.toString().split('\n').filter(Boolean).forEach(line => {
+            if (line.trim() === 'READY') {
                 isReady = true;
-                console.log('Inference worker is ready.');
-                continue;
+                console.log('--- Model fully loaded and ready ---');
+                return;
             }
             try {
                 const parsed = JSON.parse(line);
-                const resolverCount = Object.keys(resolvers).length;
-                if (resolverCount > 0) {
-                    const firstKey = Object.keys(resolvers)[0];
+                const firstKey = Object.keys(resolvers)[0];
+                if (firstKey) {
                     resolvers[firstKey](parsed);
+                    delete resolvers[firstKey];
                 }
             }
-            catch (e) {
-                console.log('Got non-JSON output from worker:', line);
-            }
-        }
-    });
-    inferProcess.on('exit', (code) => {
-        console.log(`Inference worker exited with code ${code}`);
-        process.exit(code || 1);
+            catch { }
+        });
     });
-}).catch(e => {
-    console.error('Failed to download models:', e);
-    process.exit(1);
-});
-// ── Request queue ─────────────────────────────────────────────────────
+    inferProcess.on('exit', (code) => process.exit(code || 1));
+}).catch(e => { console.error(e); process.exit(1); });
 const requestQueue = [];
 let isProcessing = false;
 function processQueue() {
@@ -226,42 +239,24 @@ function processQueue() {
         return;
     isProcessing = true;
     const req = requestQueue.shift();
-    console.log(`  [Queue]   Processing request #${req.reqId} (queue length: ${requestQueue.length})`);
     resolvers[req.file] = (result) => {
-        delete resolvers[req.file];
         isProcessing = false;
-        if (fs_1.default.existsSync(req.file)) {
+        if (fs_1.default.existsSync(req.file))
             fs_1.default.unlinkSync(req.file);
-        }
         req.resolve(result);
-        process.nextTick(processQueue);
+        processQueue();
     };
     inferProcess.stdin.write(req.file + '\n');
 }
-// ── Transcription endpoint ────────────────────────────────────────────
-app.post('/transcribe', express_1.default.raw({ type: 'audio/wav', limit: '50mb' }), async (req, res) => {
-    const reqId = req._reqId || 0;
-    if (!isReady) {
-        console.log(`  [#${reqId}]  Rejected: models still loading`);
-        return res.status(503).json({ error: 'Models are still loading' });
-    }
-    if (!req.body || !Buffer.isBuffer(req.body)) {
-        console.log(`  [#${reqId}]  Rejected: invalid audio body`);
-        return res.status(400).json({ error: 'Invalid audio body. Send raw WAV bytes with Content-Type: audio/wav' });
-    }
-    const audioSize = req.body.length;
-    console.log(`  [#${reqId}]  Audio received: ${formatBytes(audioSize)}`);
-    const tempFilePath = path_1.default.join(uploadDir, `upload-${Date.now()}-${Math.random().toString(36).substring(7)}.wav`);
-    fs_1.default.writeFileSync(tempFilePath, req.body);
-    console.log(`  [#${reqId}]  Queued for inference (queue length: ${requestQueue.length})`);
-    const result = await new Promise((resolve) => {
-        requestQueue.push({ file: tempFilePath, resolve, reqId });
+app.post('/transcribe', express_1.default.raw({ type: 'audio/wav', limit: '100mb' }), async (req, res) => {
+    if (!isReady)
+        return res.status(503).json({ error: 'Starting up' });
+    const tempFile = path_1.default.join(uploadDir, `up-${Date.now()}.wav`);
+    fs_1.default.writeFileSync(tempFile, req.body);
+    const result = await new Promise(r => {
+        requestQueue.push({ file: tempFile, resolve: r, reqId: req._reqId });
         processQueue();
     });
     res.json(result);
 });
-// ── Start server ──────────────────────────────────────────────────────
-app.listen(port, () => {
-    console.log(`\nHandy Remote Server is running on port ${port}`);
-    console.log(`Waiting for requests...\n`);
-});
+app.listen(port, () => console.log(`\nHandy Server on port ${port} | API Key: ${API_KEY}`));

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "handy-remote-server",
-  "version": "1.2.0",
+  "version": "1.3.0",
   "description": "Remote Transcription Server for Handy",
   "main": "dist/index.js",
   "bin": {
@@ -34,9 +34,13 @@
   },
   "devDependencies": {
     "@types/express": "^5.0.6",
+    "@types/gunzip-maybe": "^1.4.3",
     "@types/multer": "^2.1.0",
     "@types/node": "^25.3.5",
+    "@types/tar-fs": "^2.0.4",
+    "gunzip-maybe": "^1.4.2",
+    "tar-fs": "^3.1.2",
     "ts-node": "^10.9.2",
     "typescript": "^5.9.3"
   }
-}
+}

package/rust-infer/src/main.rs

@@ -1,35 +1,91 @@
 use std::io::{self, BufRead, Write};
-use std::path::PathBuf;
-use transcribe_rs::{engines::gigaam::GigaAMEngine, TranscriptionEngine};
+use std::path::{Path, PathBuf};
+use transcribe_rs::TranscriptionEngine;
+use transcribe_rs::engines::{
+    gigaam::GigaAMEngine,
+    whisper::WhisperEngine,
+    moonshine::{MoonshineEngine, MoonshineModelParams, ModelVariant},
+    parakeet::{ParakeetEngine, ParakeetModelParams},
+    sense_voice::{SenseVoiceEngine, SenseVoiceModelParams},
+};
+
+enum EngineWrapper {
+    GigaAM(GigaAMEngine),
+    Whisper(WhisperEngine),
+    Moonshine(MoonshineEngine),
+    Parakeet(ParakeetEngine),
+    SenseVoice(SenseVoiceEngine),
+}
+
+impl EngineWrapper {
+    fn transcribe_samples(&mut self, audio: Vec<f32>) -> Result<transcribe_rs::TranscriptionResult, Box<dyn std::error::Error>> {
+        match self {
+            EngineWrapper::GigaAM(e) => e.transcribe_samples(audio, None).map_err(|e| e.into()),
+            EngineWrapper::Whisper(e) => e.transcribe_samples(audio, None).map_err(|e| e.into()),
+            EngineWrapper::Moonshine(e) => e.transcribe_samples(audio, None).map_err(|e| e.into()),
+            EngineWrapper::Parakeet(e) => e.transcribe_samples(audio, None).map_err(|e| e.into()),
+            EngineWrapper::SenseVoice(e) => e.transcribe_samples(audio, None).map_err(|e| e.into()),
+        }
+    }
+}
 
 fn main() -> Result<(), Box<dyn std::error::Error>> {
-    // We get model and config path from args or default
     let args: Vec<String> = std::env::args().collect();
     
-    let model_path = if args.len() > 1 {
-        args[1].clone()
-    } else {
-        let mut d = std::env::current_dir()?;
-        d.push("models");
-        d.push("gigaam.onnx");
-        d.to_string_lossy().to_string()
-    };
-    
-    // Auto-download logic or print error if missing
-    if !PathBuf::from(&model_path).exists() {
-        eprintln!("Model file not found: {}. Please ensure model file exists.", model_path);
+    // Usage: rust-infer <engine_type> <model_path>
+    if args.len() < 3 {
+        eprintln!("Usage: rust-infer <engine_type> <model_path>");
+        eprintln!("Engines: gigaam, whisper, moonshine, parakeet, sensevoice");
         std::process::exit(1);
     }
-    
-    eprintln!("Loading GigaAM model from {}...", model_path);
-    
-    let mut engine = GigaAMEngine::new();
-    engine.load_model(std::path::Path::new(&model_path)).map_err(|e| format!("Failed to load GigaAM model: {}", e))?;
-    
+
+    let engine_type = args[1].to_lowercase();
+    let model_path = &args[2];
+
+    if !PathBuf::from(model_path).exists() {
+        eprintln!("Model file not found: {}", model_path);
+        std::process::exit(1);
+    }
+
+    eprintln!("Loading {} engine with model {}...", engine_type, model_path);
+
+    let mut engine = match engine_type.as_str() {
+        "gigaam" => {
+            let mut e = GigaAMEngine::new();
+            e.load_model(Path::new(model_path))?;
+            EngineWrapper::GigaAM(e)
+        }
+        "whisper" => {
+            let mut e = WhisperEngine::new();
+            e.load_model(Path::new(model_path))?;
+            EngineWrapper::Whisper(e)
+        }
+        "moonshine" => {
+            let mut e = MoonshineEngine::new();
+            // Use Base as default for remote
+            e.load_model_with_params(Path::new(model_path), MoonshineModelParams::variant(ModelVariant::Base))?;
+            EngineWrapper::Moonshine(e)
+        }
+        "parakeet" => {
+            let mut e = ParakeetEngine::new();
+            e.load_model_with_params(Path::new(model_path), ParakeetModelParams::int8())?;
+            EngineWrapper::Parakeet(e)
+        }
+        "sensevoice" => {
+            let mut e = SenseVoiceEngine::new();
+            e.load_model_with_params(Path::new(model_path), SenseVoiceModelParams::int8())?;
+            EngineWrapper::SenseVoice(e)
+        }
+        _ => {
+            eprintln!("Unknown engine type: {}", engine_type);
+            std::process::exit(1);
+        }
+    };
+
     eprintln!("Model loaded. Ready to transcribe.");
-    println!("READY"); // Signal to Node.js that we are ready
+    println!("READY");
     io::stdout().flush()?;
-    
+
     let stdin = io::stdin();
     for line in stdin.lock().lines() {
         let line = line?;
@@ -37,64 +93,47 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
         if line.is_empty() {
             continue;
         }
-        
         if line == "EXIT" {
             break;
         }
-        
-        // Line format: "file_path"
-        let wav_path = line;
-        
-        // Read the file and convert to f32
-        match read_wav(wav_path) {
+
+        match read_wav(line) {
             Ok(samples) => {
-                match engine.transcribe_samples(samples, None) {
+                match engine.transcribe_samples(samples) {
                     Ok(result) => {
-                        let json = serde_json::json!({
-                            "status": "success",
-                            "text": result.text
-                        });
+                        let json = serde_json::json!({ "status": "success", "text": result.text });
                         println!("{}", json.to_string());
-                    },
+                    }
                     Err(e) => {
-                        let json = serde_json::json!({
-                            "status": "error",
-                            "error": format!("Transcription failed: {}", e)
-                        });
+                        let json = serde_json::json!({ "status": "error", "error": format!("Transcription failed: {}", e) });
                         println!("{}", json.to_string());
                     }
                 }
-            },
+            }
             Err(e) => {
-                let json = serde_json::json!({
-                    "status": "error",
-                    "error": format!("Failed to read WAV: {}", e)
-                });
+                let json = serde_json::json!({ "status": "error", "error": format!("Failed to read WAV: {}", e) });
                 println!("{}", json.to_string());
             }
         }
         io::stdout().flush()?;
     }
-    
     Ok(())
 }
 
 fn read_wav(path: &str) -> Result<Vec<f32>, Box<dyn std::error::Error>> {
     let mut reader = hound::WavReader::open(path)?;
     let spec = reader.spec();
-    
     let mut samples = Vec::new();
     match spec.sample_format {
         hound::SampleFormat::Int => {
             if spec.bits_per_sample == 16 {
                 for sample in reader.samples::<i16>() {
-                    let s = sample? as f32 / i16::MAX as f32;
-                    samples.push(s);
+                    samples.push(sample? as f32 / i16::MAX as f32);
                 }
             } else {
                 return Err("Only 16-bit integer WAV is supported".into());
             }
-        },
+        }
         hound::SampleFormat::Float => {
             if spec.bits_per_sample == 32 {
                 for sample in reader.samples::<f32>() {
@@ -105,17 +144,13 @@ fn read_wav(path: &str) -> Result<Vec<f32>, Box<dyn std::error::Error>> {
             }
         }
     }
-    
-    // Multi-channel to mono (simple average)
     if spec.channels > 1 {
         let channels = spec.channels as usize;
         let mut mono = Vec::with_capacity(samples.len() / channels);
         for chunk in samples.chunks(channels) {
-            let sum: f32 = chunk.iter().sum();
-            mono.push(sum / channels as f32);
+            mono.push(chunk.iter().sum::<f32>() / channels as f32);
         }
         samples = mono;
     }
-    
     Ok(samples)
 }