whisper-cpp-node 0.2.4 → 0.2.5

package/README.md

@@ -1,353 +1,353 @@
-# whisper-cpp-node
-
-Node.js bindings for [whisper.cpp](https://github.com/ggerganov/whisper.cpp) - fast speech-to-text with GPU acceleration.
-
-## Features
-
-- **Fast**: Native whisper.cpp performance with GPU acceleration
-- **Cross-platform**: macOS (Metal), Windows (Vulkan)
-- **Core ML**: Optional Apple Neural Engine support for 3x+ speedup (macOS)
-- **OpenVINO**: Optional Intel CPU/GPU encoder acceleration (Windows/Linux)
-- **Streaming VAD**: Built-in Silero voice activity detection
-- **TypeScript**: Full type definitions included
-- **Self-contained**: No external dependencies, just install and use
-
-## Requirements
-
-**macOS:**
-- macOS 13.3+ (Ventura or later)
-- Apple Silicon (M1/M2/M3/M4)
-- Node.js 18+
-
-**Windows:**
-- Windows 10/11 (x64)
-- Node.js 18+
-- Vulkan-capable GPU (optional, for GPU acceleration)
-
-## Installation
-
-```bash
-npm install whisper-cpp-node
-# or
-pnpm add whisper-cpp-node
-```
-
-The platform-specific binary is automatically installed:
-- macOS ARM64: `@whisper-cpp-node/darwin-arm64`
-- Windows x64: `@whisper-cpp-node/win32-x64`
-
-## Quick Start
-
-### File-based transcription
-
-```typescript
-import {
-  createWhisperContext,
-  transcribeAsync,
-} from "whisper-cpp-node";
-
-// Create a context with your model
-const ctx = createWhisperContext({
-  model: "./models/ggml-base.en.bin",
-  use_gpu: true,
-});
-
-// Transcribe audio file
-const result = await transcribeAsync(ctx, {
-  fname_inp: "./audio.wav",
-  language: "en",
-});
-
-// Result: { segments: [["00:00:00,000", "00:00:02,500", " Hello world"], ...] }
-for (const [start, end, text] of result.segments) {
-  console.log(`[${start} --> ${end}]${text}`);
-}
-
-// Clean up
-ctx.free();
-```
-
-### Buffer-based transcription
-
-```typescript
-import {
-  createWhisperContext,
-  transcribeAsync,
-} from "whisper-cpp-node";
-
-const ctx = createWhisperContext({
-  model: "./models/ggml-base.en.bin",
-  use_gpu: true,
-});
-
-// Pass raw PCM audio (16kHz, mono, float32)
-const pcmData = new Float32Array(/* your audio samples */);
-const result = await transcribeAsync(ctx, {
-  pcmf32: pcmData,
-  language: "en",
-});
-
-for (const [start, end, text] of result.segments) {
-  console.log(`[${start} --> ${end}]${text}`);
-}
-
-ctx.free();
-```
-
-### Streaming transcription
-
-Get real-time output as audio is processed. The `on_new_segment` callback fires for each segment as it's generated, while the final callback still receives all segments at completion (backward compatible):
-
-```typescript
-import { createWhisperContext, transcribe } from "whisper-cpp-node";
-
-const ctx = createWhisperContext({
-  model: "./models/ggml-base.en.bin",
-});
-
-transcribe(ctx, {
-  fname_inp: "./long-audio.wav",
-  language: "en",
-
-  // Called for each segment as it's generated
-  on_new_segment: (segment) => {
-    console.log(`[${segment.start}]${segment.text}`);
-  },
-}, (err, result) => {
-  // Final callback still receives ALL segments at completion
-  console.log(`Done! ${result.segments.length} segments`);
-  ctx.free();
-});
-```
-
-## API
-
-### `createWhisperContext(options)`
-
-Create a persistent context for transcription.
-
-```typescript
-interface WhisperContextOptions {
-  model: string;           // Path to GGML model file (required)
-  use_gpu?: boolean;       // Enable GPU acceleration (default: true)
-                           // Uses Metal on macOS, Vulkan on Windows
-  use_coreml?: boolean;    // Enable Core ML on macOS (default: false)
-  use_openvino?: boolean;  // Enable OpenVINO encoder on Intel (default: false)
-  openvino_device?: string; // OpenVINO device: 'CPU', 'GPU', 'NPU' (default: 'CPU')
-  openvino_model_path?: string; // Path to OpenVINO encoder model (auto-derived)
-  openvino_cache_dir?: string;  // Cache dir for compiled OpenVINO models
-  flash_attn?: boolean;    // Enable Flash Attention (default: false)
-  gpu_device?: number;     // GPU device index (default: 0)
-  dtw?: string;            // DTW preset for word timestamps
-  no_prints?: boolean;     // Suppress log output (default: false)
-}
-```
-
-### `transcribeAsync(context, options)`
-
-Transcribe audio (Promise-based). Accepts either a file path or PCM buffer.
-
-```typescript
-// File input
-interface TranscribeOptionsFile {
-  fname_inp: string;       // Path to audio file
-  // ... common options
-}
-
-// Buffer input
-interface TranscribeOptionsBuffer {
-  pcmf32: Float32Array;    // Raw PCM (16kHz, mono, float32, -1.0 to 1.0)
-  // ... common options
-}
-
-// Common options (partial list - see types.ts for full options)
-interface TranscribeOptionsBase {
-  // Language
-  language?: string;       // Language code ('en', 'zh', 'auto')
-  translate?: boolean;     // Translate to English
-  detect_language?: boolean; // Auto-detect language
-
-  // Threading
-  n_threads?: number;      // CPU threads (default: 4)
-  n_processors?: number;   // Parallel processors
-
-  // Audio processing
-  offset_ms?: number;      // Start offset in ms
-  duration_ms?: number;    // Duration to process (0 = all)
-
-  // Output control
-  no_timestamps?: boolean; // Disable timestamps
-  max_len?: number;        // Max segment length (chars)
-  max_tokens?: number;     // Max tokens per segment
-  split_on_word?: boolean; // Split on word boundaries
-  token_timestamps?: boolean; // Include token-level timestamps
-
-  // Sampling
-  temperature?: number;    // Sampling temperature (0.0 = greedy)
-  beam_size?: number;      // Beam search size (-1 = greedy)
-  best_of?: number;        // Best-of-N sampling
-
-  // Thresholds
-  entropy_thold?: number;  // Entropy threshold
-  logprob_thold?: number;  // Log probability threshold
-  no_speech_thold?: number; // No-speech probability threshold
-
-  // Context
-  prompt?: string;         // Initial prompt text
-  no_context?: boolean;    // Don't use previous context
-
-  // VAD preprocessing
-  vad?: boolean;           // Enable VAD preprocessing
-  vad_model?: string;      // Path to VAD model
-  vad_threshold?: number;  // VAD threshold (0.0-1.0)
-  vad_min_speech_duration_ms?: number;
-  vad_min_silence_duration_ms?: number;
-  vad_speech_pad_ms?: number;
-
-  // Callbacks
-  progress_callback?: (progress: number) => void;
-  on_new_segment?: (segment: StreamingSegment) => void;  // Streaming callback
-}
-
-// Streaming segment (passed to on_new_segment callback)
-interface StreamingSegment {
-  start: string;           // Start timestamp "HH:MM:SS,mmm"
-  end: string;             // End timestamp
-  text: string;            // Transcribed text
-  segment_index: number;   // 0-based index
-  is_partial: boolean;     // Reserved for future use
-  tokens?: StreamingToken[]; // Only if token_timestamps enabled
-}
-
-// Result
-interface TranscribeResult {
-  segments: TranscriptSegment[];
-}
-
-// Segment is a tuple: [start, end, text]
-type TranscriptSegment = [string, string, string];
-// Example: ["00:00:00,000", "00:00:02,500", " Hello world"]
-```
-
-### `createVadContext(options)`
-
-Create a voice activity detection context for streaming audio.
-
-```typescript
-interface VadContextOptions {
-  model: string;           // Path to Silero VAD model
-  threshold?: number;      // Speech threshold (default: 0.5)
-  n_threads?: number;      // Number of threads (default: 1)
-  no_prints?: boolean;     // Suppress log output
-}
-
-interface VadContext {
-  getWindowSamples(): number;  // Returns 512 (32ms at 16kHz)
-  getSampleRate(): number;     // Returns 16000
-  process(samples: Float32Array): number;  // Returns probability 0.0-1.0
-  reset(): void;               // Reset LSTM state
-  free(): void;                // Release resources
-}
-```
-
-#### VAD Example
-
-```typescript
-import { createVadContext } from "whisper-cpp-node";
-
-const vad = createVadContext({
-  model: "./models/ggml-silero-v6.2.0.bin",
-  threshold: 0.5,
-});
-
-const windowSize = vad.getWindowSamples(); // 512 samples
-
-// Process audio in 32ms chunks
-function processAudioChunk(samples: Float32Array) {
-  const probability = vad.process(samples);
-  if (probability >= 0.5) {
-    console.log("Speech detected!", probability);
-  }
-}
-
-// Reset when starting new audio stream
-vad.reset();
-
-// Clean up when done
-vad.free();
-```
-
-## Core ML Acceleration (macOS)
-
-For 3x+ faster encoding on Apple Silicon:
-
-1. Generate a Core ML model:
-   ```bash
-   pip install ane_transformers openai-whisper coremltools
-   ./models/generate-coreml-model.sh base.en
-   ```
-
-2. Place it next to your GGML model:
-   ```
-   models/ggml-base.en.bin
-   models/ggml-base.en-encoder.mlmodelc/
-   ```
-
-3. Enable Core ML:
-   ```typescript
-   const ctx = createWhisperContext({
-     model: "./models/ggml-base.en.bin",
-     use_coreml: true,
-   });
-   ```
-
-## OpenVINO Acceleration (Intel)
-
-For faster encoder inference on Intel CPUs and GPUs (requires build with OpenVINO support):
-
-1. Install OpenVINO and convert the model:
-   ```bash
-   pip install openvino openvino-dev
-   python models/convert-whisper-to-openvino.py --model base.en
-   ```
-
-2. The OpenVINO model files are placed next to your GGML model:
-   ```
-   models/ggml-base.en.bin
-   models/ggml-base.en-encoder-openvino.xml
-   models/ggml-base.en-encoder-openvino.bin
-   ```
-
-3. Enable OpenVINO:
-   ```typescript
-   const ctx = createWhisperContext({
-     model: "./models/ggml-base.en.bin",
-     use_openvino: true,
-     openvino_device: "CPU",  // or "GPU" for Intel iGPU
-     openvino_cache_dir: "./openvino_cache", // optional, speeds up init
-   });
-   ```
-
-**Note:** OpenVINO support requires the addon to be built with `-DADDON_OPENVINO=ON`.
-
-## Models
-
-Download models from [Hugging Face](https://huggingface.co/ggerganov/whisper.cpp):
-
-```bash
-# Base English model (~150MB)
-curl -L -o models/ggml-base.en.bin \
-  https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en.bin
-
-# Large v3 Turbo quantized (~500MB)
-curl -L -o models/ggml-large-v3-turbo-q4_0.bin \
-  https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-large-v3-turbo-q4_0.bin
-
-# Silero VAD model (for streaming VAD)
-curl -L -o models/ggml-silero-v6.2.0.bin \
-  https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-silero-v6.2.0.bin
-```
-
-## License
-
-MIT
+# whisper-cpp-node
+
+Node.js bindings for [whisper.cpp](https://github.com/ggerganov/whisper.cpp) - fast speech-to-text with GPU acceleration.
+
+## Features
+
+- **Fast**: Native whisper.cpp performance with GPU acceleration
+- **Cross-platform**: macOS (Metal), Windows (Vulkan)
+- **Core ML**: Optional Apple Neural Engine support for 3x+ speedup (macOS)
+- **OpenVINO**: Optional Intel CPU/GPU encoder acceleration (Windows/Linux)
+- **Streaming VAD**: Built-in Silero voice activity detection
+- **TypeScript**: Full type definitions included
+- **Self-contained**: No external dependencies, just install and use
+
+## Requirements
+
+**macOS:**
+- macOS 13.3+ (Ventura or later)
+- Apple Silicon (M1/M2/M3/M4)
+- Node.js 18+
+
+**Windows:**
+- Windows 10/11 (x64)
+- Node.js 18+
+- Vulkan-capable GPU (optional, for GPU acceleration)
+
+## Installation
+
+```bash
+npm install whisper-cpp-node
+# or
+pnpm add whisper-cpp-node
+```
+
+The platform-specific binary is automatically installed:
+- macOS ARM64: `@whisper-cpp-node/darwin-arm64`
+- Windows x64: `@whisper-cpp-node/win32-x64`
+
+## Quick Start
+
+### File-based transcription
+
+```typescript
+import {
+  createWhisperContext,
+  transcribeAsync,
+} from "whisper-cpp-node";
+
+// Create a context with your model
+const ctx = createWhisperContext({
+  model: "./models/ggml-base.en.bin",
+  use_gpu: true,
+});
+
+// Transcribe audio file
+const result = await transcribeAsync(ctx, {
+  fname_inp: "./audio.wav",
+  language: "en",
+});
+
+// Result: { segments: [["00:00:00,000", "00:00:02,500", " Hello world"], ...] }
+for (const [start, end, text] of result.segments) {
+  console.log(`[${start} --> ${end}]${text}`);
+}
+
+// Clean up
+ctx.free();
+```
+
+### Buffer-based transcription
+
+```typescript
+import {
+  createWhisperContext,
+  transcribeAsync,
+} from "whisper-cpp-node";
+
+const ctx = createWhisperContext({
+  model: "./models/ggml-base.en.bin",
+  use_gpu: true,
+});
+
+// Pass raw PCM audio (16kHz, mono, float32)
+const pcmData = new Float32Array(/* your audio samples */);
+const result = await transcribeAsync(ctx, {
+  pcmf32: pcmData,
+  language: "en",
+});
+
+for (const [start, end, text] of result.segments) {
+  console.log(`[${start} --> ${end}]${text}`);
+}
+
+ctx.free();
+```
+
+### Streaming transcription
+
+Get real-time output as audio is processed. The `on_new_segment` callback fires for each segment as it's generated, while the final callback still receives all segments at completion (backward compatible):
+
+```typescript
+import { createWhisperContext, transcribe } from "whisper-cpp-node";
+
+const ctx = createWhisperContext({
+  model: "./models/ggml-base.en.bin",
+});
+
+transcribe(ctx, {
+  fname_inp: "./long-audio.wav",
+  language: "en",
+
+  // Called for each segment as it's generated
+  on_new_segment: (segment) => {
+    console.log(`[${segment.start}]${segment.text}`);
+  },
+}, (err, result) => {
+  // Final callback still receives ALL segments at completion
+  console.log(`Done! ${result.segments.length} segments`);
+  ctx.free();
+});
+```
+
+## API
+
+### `createWhisperContext(options)`
+
+Create a persistent context for transcription.
+
+```typescript
+interface WhisperContextOptions {
+  model: string;           // Path to GGML model file (required)
+  use_gpu?: boolean;       // Enable GPU acceleration (default: true)
+                           // Uses Metal on macOS, Vulkan on Windows
+  use_coreml?: boolean;    // Enable Core ML on macOS (default: false)
+  use_openvino?: boolean;  // Enable OpenVINO encoder on Intel (default: false)
+  openvino_device?: string; // OpenVINO device: 'CPU', 'GPU', 'NPU' (default: 'CPU')
+  openvino_model_path?: string; // Path to OpenVINO encoder model (auto-derived)
+  openvino_cache_dir?: string;  // Cache dir for compiled OpenVINO models
+  flash_attn?: boolean;    // Enable Flash Attention (default: false)
+  gpu_device?: number;     // GPU device index (default: 0)
+  dtw?: string;            // DTW preset for word timestamps
+  no_prints?: boolean;     // Suppress log output (default: false)
+}
+```
+
+### `transcribeAsync(context, options)`
+
+Transcribe audio (Promise-based). Accepts either a file path or PCM buffer.
+
+```typescript
+// File input
+interface TranscribeOptionsFile {
+  fname_inp: string;       // Path to audio file
+  // ... common options
+}
+
+// Buffer input
+interface TranscribeOptionsBuffer {
+  pcmf32: Float32Array;    // Raw PCM (16kHz, mono, float32, -1.0 to 1.0)
+  // ... common options
+}
+
+// Common options (partial list - see types.ts for full options)
+interface TranscribeOptionsBase {
+  // Language
+  language?: string;       // Language code ('en', 'zh', 'auto')
+  translate?: boolean;     // Translate to English
+  detect_language?: boolean; // Auto-detect language
+
+  // Threading
+  n_threads?: number;      // CPU threads (default: 4)
+  n_processors?: number;   // Parallel processors
+
+  // Audio processing
+  offset_ms?: number;      // Start offset in ms
+  duration_ms?: number;    // Duration to process (0 = all)
+
+  // Output control
+  no_timestamps?: boolean; // Disable timestamps
+  max_len?: number;        // Max segment length (chars)
+  max_tokens?: number;     // Max tokens per segment
+  split_on_word?: boolean; // Split on word boundaries
+  token_timestamps?: boolean; // Include token-level timestamps
+
+  // Sampling
+  temperature?: number;    // Sampling temperature (0.0 = greedy)
+  beam_size?: number;      // Beam search size (-1 = greedy)
+  best_of?: number;        // Best-of-N sampling
+
+  // Thresholds
+  entropy_thold?: number;  // Entropy threshold
+  logprob_thold?: number;  // Log probability threshold
+  no_speech_thold?: number; // No-speech probability threshold
+
+  // Context
+  prompt?: string;         // Initial prompt text
+  no_context?: boolean;    // Don't use previous context
+
+  // VAD preprocessing
+  vad?: boolean;           // Enable VAD preprocessing
+  vad_model?: string;      // Path to VAD model
+  vad_threshold?: number;  // VAD threshold (0.0-1.0)
+  vad_min_speech_duration_ms?: number;
+  vad_min_silence_duration_ms?: number;
+  vad_speech_pad_ms?: number;
+
+  // Callbacks
+  progress_callback?: (progress: number) => void;
+  on_new_segment?: (segment: StreamingSegment) => void;  // Streaming callback
+}
+
+// Streaming segment (passed to on_new_segment callback)
+interface StreamingSegment {
+  start: string;           // Start timestamp "HH:MM:SS,mmm"
+  end: string;             // End timestamp
+  text: string;            // Transcribed text
+  segment_index: number;   // 0-based index
+  is_partial: boolean;     // Reserved for future use
+  tokens?: StreamingToken[]; // Only if token_timestamps enabled
+}
+
+// Result
+interface TranscribeResult {
+  segments: TranscriptSegment[];
+}
+
+// Segment is a tuple: [start, end, text]
+type TranscriptSegment = [string, string, string];
+// Example: ["00:00:00,000", "00:00:02,500", " Hello world"]
+```
+
+### `createVadContext(options)`
+
+Create a voice activity detection context for streaming audio.
+
+```typescript
+interface VadContextOptions {
+  model: string;           // Path to Silero VAD model
+  threshold?: number;      // Speech threshold (default: 0.5)
+  n_threads?: number;      // Number of threads (default: 1)
+  no_prints?: boolean;     // Suppress log output
+}
+
+interface VadContext {
+  getWindowSamples(): number;  // Returns 512 (32ms at 16kHz)
+  getSampleRate(): number;     // Returns 16000
+  process(samples: Float32Array): number;  // Returns probability 0.0-1.0
+  reset(): void;               // Reset LSTM state
+  free(): void;                // Release resources
+}
+```
+
+#### VAD Example
+
+```typescript
+import { createVadContext } from "whisper-cpp-node";
+
+const vad = createVadContext({
+  model: "./models/ggml-silero-v6.2.0.bin",
+  threshold: 0.5,
+});
+
+const windowSize = vad.getWindowSamples(); // 512 samples
+
+// Process audio in 32ms chunks
+function processAudioChunk(samples: Float32Array) {
+  const probability = vad.process(samples);
+  if (probability >= 0.5) {
+    console.log("Speech detected!", probability);
+  }
+}
+
+// Reset when starting new audio stream
+vad.reset();
+
+// Clean up when done
+vad.free();
+```
+
+## Core ML Acceleration (macOS)
+
+For 3x+ faster encoding on Apple Silicon:
+
+1. Generate a Core ML model:
+   ```bash
+   pip install ane_transformers openai-whisper coremltools
+   ./models/generate-coreml-model.sh base.en
+   ```
+
+2. Place it next to your GGML model:
+   ```
+   models/ggml-base.en.bin
+   models/ggml-base.en-encoder.mlmodelc/
+   ```
+
+3. Enable Core ML:
+   ```typescript
+   const ctx = createWhisperContext({
+     model: "./models/ggml-base.en.bin",
+     use_coreml: true,
+   });
+   ```
+
+## OpenVINO Acceleration (Intel)
+
+For faster encoder inference on Intel CPUs and GPUs (requires build with OpenVINO support):
+
+1. Install OpenVINO and convert the model:
+   ```bash
+   pip install openvino openvino-dev
+   python models/convert-whisper-to-openvino.py --model base.en
+   ```
+
+2. The OpenVINO model files are placed next to your GGML model:
+   ```
+   models/ggml-base.en.bin
+   models/ggml-base.en-encoder-openvino.xml
+   models/ggml-base.en-encoder-openvino.bin
+   ```
+
+3. Enable OpenVINO:
+   ```typescript
+   const ctx = createWhisperContext({
+     model: "./models/ggml-base.en.bin",
+     use_openvino: true,
+     openvino_device: "CPU",  // or "GPU" for Intel iGPU
+     openvino_cache_dir: "./openvino_cache", // optional, speeds up init
+   });
+   ```
+
+**Note:** OpenVINO support requires the addon to be built with `-DADDON_OPENVINO=ON`.
+
+## Models
+
+Download models from [Hugging Face](https://huggingface.co/ggerganov/whisper.cpp):
+
+```bash
+# Base English model (~150MB)
+curl -L -o models/ggml-base.en.bin \
+  https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en.bin
+
+# Large v3 Turbo quantized (~500MB)
+curl -L -o models/ggml-large-v3-turbo-q4_0.bin \
+  https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-large-v3-turbo-q4_0.bin
+
+# Silero VAD model (for streaming VAD)
+curl -L -o models/ggml-silero-v6.2.0.bin \
+  https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-silero-v6.2.0.bin
+```
+
+## License
+
+MIT

package/dist/types.d.ts

@@ -34,8 +34,27 @@ export interface WhisperContextOptions {
      * Can speed up init time, especially for GPU, by caching compiled 'blobs'
      */
     openvino_cache_dir?: string;
-    /** DTW alignment preset for word-level timestamps (e.g., 'base.en', 'small', 'large.v3') */
+    /**
+     * DTW alignment preset for word-level timestamps.
+     *
+     * Named presets: 'tiny', 'tiny.en', 'base', 'base.en', 'small', 'small.en',
+     *   'medium', 'medium.en', 'large.v1', 'large.v2', 'large.v3', 'large.v3.turbo'
+     *
+     * Dynamic presets:
+     *   'top-N'      - Use all heads from top N text layers (plain averaging)
+     *   'top-N-norm' - Use top N layers + L2 norm filtering (keeps dtw_norm_top_k best heads)
+     *
+     * @example
+     * dtw: 'large.v3.turbo'   // named preset
+     * dtw: 'top-4'            // top 4 layers, all heads averaged
+     * dtw: 'top-4-norm'       // top 4 layers, L2 norm selects best heads
+     */
     dtw?: string;
+    /**
+     * Number of attention heads to keep after L2 norm filtering (default: 10).
+     * Only used when dtw is set to a 'top-N-norm' preset.
+     */
+    dtw_norm_top_k?: number;
     /** Suppress whisper.cpp log output (default: false) */
     no_prints?: boolean;
 }
@@ -159,12 +178,18 @@ export interface TranscribeOptionsBuffer extends TranscribeOptionsBase {
  */
 export type TranscribeOptions = TranscribeOptionsFile | TranscribeOptionsBuffer;
 /**
- * Transcription result segment (tuple format)
- * [0]: Start time in format "HH:MM:SS,mmm"
- * [1]: End time in format "HH:MM:SS,mmm"
- * [2]: Transcribed text
+ * Transcription result segment
  */
-export type TranscriptSegment = [start: string, end: string, text: string];
+export interface TranscriptSegment {
+    /** Start time in format "HH:MM:SS,mmm" */
+    start: string;
+    /** End time in format "HH:MM:SS,mmm" */
+    end: string;
+    /** Transcribed text */
+    text: string;
+    /** Token-level data (only present when token_timestamps is enabled) */
+    tokens?: StreamingToken[];
+}
 /**
  * Token information for streaming callbacks
  */
@@ -173,10 +198,16 @@ export interface StreamingToken {
     text: string;
     /** Token probability (0.0 to 1.0) */
     probability: number;
-    /** Token timestamp start (in centiseconds from audio start, only if token_timestamps enabled) */
-    t0?: number;
-    /** Token timestamp end (in centiseconds from audio start, only if token_timestamps enabled) */
-    t1?: number;
+    /** Token timestamp start (in centiseconds from audio start) */
+    t0: number;
+    /** Token timestamp end (in centiseconds from audio start) */
+    t1: number;
+    /**
+     * DTW-aligned timestamp (in centiseconds from audio start).
+     * Only meaningful when the context was created with a DTW preset.
+     * Typically more accurate than t0/t1 for word-level alignment.
+     */
+    t_dtw: number;
 }
 /**
  * Segment data passed to streaming callback
@@ -199,8 +230,10 @@ export interface StreamingSegment {
  * Transcription result
  */
 export interface TranscribeResult {
-    /** Array of transcript segments as [start, end, text] tuples */
+    /** Array of transcript segments */
     segments: TranscriptSegment[];
+    /** Detected language (when detect_language is true) */
+    language?: string;
 }
 /**
  * Options for creating a VadContext

package/dist/types.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,MAAM,WAAW,qBAAqB;IACpC,kCAAkC;IAClC,KAAK,EAAE,MAAM,CAAC;IACd,8CAA8C;IAC9C,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,8CAA8C;IAC9C,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,oCAAoC;IACpC,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,4DAA4D;IAC5D,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB;;;;;OAKG;IACH,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB;;;OAGG;IACH,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B;;;OAGG;IACH,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB;;;OAGG;IACH,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,4FAA4F;IAC5F,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,uDAAuD;IACvD,SAAS,CAAC,EAAE,OAAO,CAAC;CACrB;AAED;;GAEG;AACH,MAAM,WAAW,qBAAqB;IAEpC,+CAA+C;IAC/C,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,2BAA2B;IAC3B,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,oCAAoC;IACpC,eAAe,CAAC,EAAE,OAAO,CAAC;IAG1B,+BAA+B;IAC/B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,mDAAmD;IACnD,YAAY,CAAC,EAAE,MAAM,CAAC;IAGtB,mCAAmC;IACnC,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,oDAAoD;IACpD,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,yBAAyB;IACzB,SAAS,CAAC,EAAE,MAAM,CAAC;IAGnB,mCAAmC;IACnC,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,0BAA0B;IAC1B,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,0DAA0D;IAC1D,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,gDAAgD;IAChD,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,0CAA0C;IAC1C,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,wCAAwC;IACxC,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,qCAAqC;IACrC,gBAAgB,CAAC,EAAE,OAAO,CAAC;IAC3B,+BAA+B;IAC/B,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,oDAAoD;IACpD,aAAa,CAAC,EAAE,OAAO,CAAC;IAGxB,8CAA8C;IAC9C,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,yCAAyC;IACzC,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,oCAAoC;IACpC,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,8CAA8C;IAC9C,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,mCAAmC;IACnC,WAAW,CAAC,EAAE,OAAO,CAAC;IAGtB,qCAAqC;IACrC,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,gCAAgC;IAChC,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,sCAAsC;IACtC,eAAe,CAAC,EAAE,MAAM,CAAC;IAGzB,sCAAsC;IACtC,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,iCAAiC;IACjC,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,6BAA6B;IAC7B,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,iCAAiC;IACjC,YAAY,CAAC,EAAE,OAAO,CAAC;IAGvB,iCAAiC;IACjC,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,oDAAoD;IACpD,WAAW,CAAC,EAAE,OAAO,CAAC;IAGtB,2BAA2B;IAC3B,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,qBAAqB;IACrB,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,4BAA4B;IAC5B,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,uBAAuB;IACvB,gBAAgB,CAAC,EAAE,OAAO,CAAC;IAG3B,+BAA+B;IAC/B,GAAG,CAAC,EAAE,OAAO,CAAC;IACd,wBAAwB;IACxB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,+CAA+C;IAC/C,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,8CAA8C;IAC9C,0BAA0B,CAAC,EAAE,MAAM,CAAC;IACpC,+CAA+C;IAC/C,2BAA2B,CAAC,EAAE,MAAM,CAAC;IACrC,yCAAyC;IACzC,yBAAyB,CAAC,EAAE,MAAM,CAAC;IACnC,qCAAqC;IACrC,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,gCAAgC;IAChC,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAG7B,mDAAmD;IACnD,iBAAiB,CAAC,EAAE,CAAC,QAAQ,EAAE,MAAM,KAAK,IAAI,CAAC;IAE/C;;;;OAIG;IACH,cAAc,CAAC,EAAE,CAAC,OAAO,EAAE,gBAAgB,KAAK,IAAI,CAAC;CACtD;AAED;;GAEG;AACH,MAAM,WAAW,qBAAsB,SAAQ,qBAAqB;IAClE,6BAA6B;IAC7B,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,CAAC,EAAE,KAAK,CAAC;CAChB;AAED;;GAEG;AACH,MAAM,WAAW,uBAAwB,SAAQ,qBAAqB;IACpE,uEAAuE;IACvE,MAAM,EAAE,YAAY,CAAC;IACrB,SAAS,CAAC,EAAE,KAAK,CAAC;CACnB;AAED;;GAEG;AACH,MAAM,MAAM,iBAAiB,GAAG,qBAAqB,GAAG,uBAAuB,CAAC;AAEhF;;;;;GAKG;AACH,MAAM,MAAM,iBAAiB,GAAG,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,CAAC;AAE3E;;GAEG;AACH,MAAM,WAAW,cAAc;IAC7B,iBAAiB;IACjB,IAAI,EAAE,MAAM,CAAC;IACb,qCAAqC;IACrC,WAAW,EAAE,MAAM,CAAC;IACpB,iGAAiG;IACjG,EAAE,CAAC,EAAE,MAAM,CAAC;IACZ,+FAA+F;IAC/F,EAAE,CAAC,EAAE,MAAM,CAAC;CACb;AAED;;GAEG;AACH,MAAM,WAAW,gBAAgB;IAC/B,0CAA0C;IAC1C,KAAK,EAAE,MAAM,CAAC;IACd,wCAAwC;IACxC,GAAG,EAAE,MAAM,CAAC;IACZ,wCAAwC;IACxC,IAAI,EAAE,MAAM,CAAC;IACb,8BAA8B;IAC9B,aAAa,EAAE,MAAM,CAAC;IACtB,yDAAyD;IACzD,UAAU,EAAE,OAAO,CAAC;IACpB,iEAAiE;IACjE,MAAM,CAAC,EAAE,cAAc,EAAE,CAAC;CAC3B;AAED;;GAEG;AACH,MAAM,WAAW,gBAAgB;IAC/B,gEAAgE;IAChE,QAAQ,EAAE,iBAAiB,EAAE,CAAC;CAC/B;AAED;;GAEG;AACH,MAAM,WAAW,iBAAiB;IAChC,wCAAwC;IACxC,KAAK,EAAE,MAAM,CAAC;IACd,gDAAgD;IAChD,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,qCAAqC;IACrC,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,oCAAoC;IACpC,SAAS,CAAC,EAAE,OAAO,CAAC;CACrB;AAED;;GAEG;AACH,MAAM,WAAW,cAAc;IAC7B,yCAAyC;IACzC,aAAa,IAAI,MAAM,CAAC;IACxB,qCAAqC;IACrC,cAAc,IAAI,OAAO,CAAC;IAC1B,6CAA6C;IAC7C,IAAI,IAAI,IAAI,CAAC;CACd;AAED;;GAEG;AACH,MAAM,WAAW,yBAAyB;IACxC,KAAK,OAAO,EAAE,qBAAqB,GAAG,cAAc,CAAC;CACtD;AAED;;GAEG;AACH,MAAM,WAAW,UAAU;IACzB,8CAA8C;IAC9C,gBAAgB,IAAI,MAAM,CAAC;IAC3B,8CAA8C;IAC9C,aAAa,IAAI,MAAM,CAAC;IACxB,iEAAiE;IACjE,OAAO,CAAC,OAAO,EAAE,YAAY,GAAG,MAAM,CAAC;IACvC,oCAAoC;IACpC,KAAK,IAAI,IAAI,CAAC;IACd,6CAA6C;IAC7C,IAAI,IAAI,IAAI,CAAC;CACd;AAED;;GAEG;AACH,MAAM,WAAW,qBAAqB;IACpC,KAAK,OAAO,EAAE,iBAAiB,GAAG,UAAU,CAAC;CAC9C;AAED;;GAEG;AACH,MAAM,MAAM,kBAAkB,GAAG,CAC/B,KAAK,EAAE,KAAK,GAAG,IAAI,EACnB,MAAM,CAAC,EAAE,gBAAgB,KACtB,IAAI,CAAC;AAEV;;GAEG;AACH,MAAM,WAAW,YAAY;IAC3B,cAAc,EAAE,yBAAyB,CAAC;IAC1C,UAAU,EAAE,qBAAqB,CAAC;IAClC,UAAU,EAAE,CACV,OAAO,EAAE,cAAc,EACvB,OAAO,EAAE,iBAAiB,EAC1B,QAAQ,EAAE,kBAAkB,KACzB,IAAI,CAAC;IACV,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CAClC"}
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,MAAM,WAAW,qBAAqB;IACpC,kCAAkC;IAClC,KAAK,EAAE,MAAM,CAAC;IACd,8CAA8C;IAC9C,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,8CAA8C;IAC9C,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,oCAAoC;IACpC,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,4DAA4D;IAC5D,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB;;;;;OAKG;IACH,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB;;;OAGG;IACH,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B;;;OAGG;IACH,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB;;;OAGG;IACH,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B;;;;;;;;;;;;;;OAcG;IACH,GAAG,CAAC,EAAE,MAAM,CAAC;IACb;;;OAGG;IACH,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,uDAAuD;IACvD,SAAS,CAAC,EAAE,OAAO,CAAC;CACrB;AAED;;GAEG;AACH,MAAM,WAAW,qBAAqB;IAEpC,+CAA+C;IAC/C,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,2BAA2B;IAC3B,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,oCAAoC;IACpC,eAAe,CAAC,EAAE,OAAO,CAAC;IAG1B,+BAA+B;IAC/B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,mDAAmD;IACnD,YAAY,CAAC,EAAE,MAAM,CAAC;IAGtB,mCAAmC;IACnC,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,oDAAoD;IACpD,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,yBAAyB;IACzB,SAAS,CAAC,EAAE,MAAM,CAAC;IAGnB,mCAAmC;IACnC,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,0BAA0B;IAC1B,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,0DAA0D;IAC1D,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,gDAAgD;IAChD,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,0CAA0C;IAC1C,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,wCAAwC;IACxC,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,qCAAqC;IACrC,gBAAgB,CAAC,EAAE,OAAO,CAAC;IAC3B,+BAA+B;IAC/B,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,oDAAoD;IACpD,aAAa,CAAC,EAAE,OAAO,CAAC;IAGxB,8CAA8C;IAC9C,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,yCAAyC;IACzC,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,oCAAoC;IACpC,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,8CAA8C;IAC9C,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,mCAAmC;IACnC,WAAW,CAAC,EAAE,OAAO,CAAC;IAGtB,qCAAqC;IACrC,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,gCAAgC;IAChC,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,sCAAsC;IACtC,eAAe,CAAC,EAAE,MAAM,CAAC;IAGzB,sCAAsC;IACtC,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,iCAAiC;IACjC,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,6BAA6B;IAC7B,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,iCAAiC;IACjC,YAAY,CAAC,EAAE,OAAO,CAAC;IAGvB,iCAAiC;IACjC,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,oDAAoD;IACpD,WAAW,CAAC,EAAE,OAAO,CAAC;IAGtB,2BAA2B;IAC3B,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,qBAAqB;IACrB,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,4BAA4B;IAC5B,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,uBAAuB;IACvB,gBAAgB,CAAC,EAAE,OAAO,CAAC;IAG3B,+BAA+B;IAC/B,GAAG,CAAC,EAAE,OAAO,CAAC;IACd,wBAAwB;IACxB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,+CAA+C;IAC/C,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,8CAA8C;IAC9C,0BAA0B,CAAC,EAAE,MAAM,CAAC;IACpC,+CAA+C;IAC/C,2BAA2B,CAAC,EAAE,MAAM,CAAC;IACrC,yCAAyC;IACzC,yBAAyB,CAAC,EAAE,MAAM,CAAC;IACnC,qCAAqC;IACrC,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,gCAAgC;IAChC,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAG7B,mDAAmD;IACnD,iBAAiB,CAAC,EAAE,CAAC,QAAQ,EAAE,MAAM,KAAK,IAAI,CAAC;IAE/C;;;;OAIG;IACH,cAAc,CAAC,EAAE,CAAC,OAAO,EAAE,gBAAgB,KAAK,IAAI,CAAC;CACtD;AAED;;GAEG;AACH,MAAM,WAAW,qBAAsB,SAAQ,qBAAqB;IAClE,6BAA6B;IAC7B,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,CAAC,EAAE,KAAK,CAAC;CAChB;AAED;;GAEG;AACH,MAAM,WAAW,uBAAwB,SAAQ,qBAAqB;IACpE,uEAAuE;IACvE,MAAM,EAAE,YAAY,CAAC;IACrB,SAAS,CAAC,EAAE,KAAK,CAAC;CACnB;AAED;;GAEG;AACH,MAAM,MAAM,iBAAiB,GAAG,qBAAqB,GAAG,uBAAuB,CAAC;AAEhF;;GAEG;AACH,MAAM,WAAW,iBAAiB;IAChC,0CAA0C;IAC1C,KAAK,EAAE,MAAM,CAAC;IACd,wCAAwC;IACxC,GAAG,EAAE,MAAM,CAAC;IACZ,uBAAuB;IACvB,IAAI,EAAE,MAAM,CAAC;IACb,uEAAuE;IACvE,MAAM,CAAC,EAAE,cAAc,EAAE,CAAC;CAC3B;AAED;;GAEG;AACH,MAAM,WAAW,cAAc;IAC7B,iBAAiB;IACjB,IAAI,EAAE,MAAM,CAAC;IACb,qCAAqC;IACrC,WAAW,EAAE,MAAM,CAAC;IACpB,+DAA+D;IAC/D,EAAE,EAAE,MAAM,CAAC;IACX,6DAA6D;IAC7D,EAAE,EAAE,MAAM,CAAC;IACX;;;;OAIG;IACH,KAAK,EAAE,MAAM,CAAC;CACf;AAED;;GAEG;AACH,MAAM,WAAW,gBAAgB;IAC/B,0CAA0C;IAC1C,KAAK,EAAE,MAAM,CAAC;IACd,wCAAwC;IACxC,GAAG,EAAE,MAAM,CAAC;IACZ,wCAAwC;IACxC,IAAI,EAAE,MAAM,CAAC;IACb,8BAA8B;IAC9B,aAAa,EAAE,MAAM,CAAC;IACtB,yDAAyD;IACzD,UAAU,EAAE,OAAO,CAAC;IACpB,iEAAiE;IACjE,MAAM,CAAC,EAAE,cAAc,EAAE,CAAC;CAC3B;AAED;;GAEG;AACH,MAAM,WAAW,gBAAgB;IAC/B,mCAAmC;IACnC,QAAQ,EAAE,iBAAiB,EAAE,CAAC;IAC9B,uDAAuD;IACvD,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAED;;GAEG;AACH,MAAM,WAAW,iBAAiB;IAChC,wCAAwC;IACxC,KAAK,EAAE,MAAM,CAAC;IACd,gDAAgD;IAChD,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,qCAAqC;IACrC,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,oCAAoC;IACpC,SAAS,CAAC,EAAE,OAAO,CAAC;CACrB;AAED;;GAEG;AACH,MAAM,WAAW,cAAc;IAC7B,yCAAyC;IACzC,aAAa,IAAI,MAAM,CAAC;IACxB,qCAAqC;IACrC,cAAc,IAAI,OAAO,CAAC;IAC1B,6CAA6C;IAC7C,IAAI,IAAI,IAAI,CAAC;CACd;AAED;;GAEG;AACH,MAAM,WAAW,yBAAyB;IACxC,KAAK,OAAO,EAAE,qBAAqB,GAAG,cAAc,CAAC;CACtD;AAED;;GAEG;AACH,MAAM,WAAW,UAAU;IACzB,8CAA8C;IAC9C,gBAAgB,IAAI,MAAM,CAAC;IAC3B,8CAA8C;IAC9C,aAAa,IAAI,MAAM,CAAC;IACxB,iEAAiE;IACjE,OAAO,CAAC,OAAO,EAAE,YAAY,GAAG,MAAM,CAAC;IACvC,oCAAoC;IACpC,KAAK,IAAI,IAAI,CAAC;IACd,6CAA6C;IAC7C,IAAI,IAAI,IAAI,CAAC;CACd;AAED;;GAEG;AACH,MAAM,WAAW,qBAAqB;IACpC,KAAK,OAAO,EAAE,iBAAiB,GAAG,UAAU,CAAC;CAC9C;AAED;;GAEG;AACH,MAAM,MAAM,kBAAkB,GAAG,CAC/B,KAAK,EAAE,KAAK,GAAG,IAAI,EACnB,MAAM,CAAC,EAAE,gBAAgB,KACtB,IAAI,CAAC;AAEV;;GAEG;AACH,MAAM,WAAW,YAAY;IAC3B,cAAc,EAAE,yBAAyB,CAAC;IAC1C,UAAU,EAAE,qBAAqB,CAAC;IAClC,UAAU,EAAE,CACV,OAAO,EAAE,cAAc,EACvB,OAAO,EAAE,iBAAiB,EAC1B,QAAQ,EAAE,kBAAkB,KACzB,IAAI,CAAC;IACV,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CAClC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "whisper-cpp-node",
-  "version": "0.2.4",
+  "version": "0.2.5",
   "description": "Node.js bindings for whisper.cpp - fast speech-to-text with GPU acceleration",
   "license": "MIT",
   "repository": {
@@ -21,8 +21,8 @@
     "dist"
   ],
   "optionalDependencies": {
-    "@whisper-cpp-node/darwin-arm64": "0.2.2",
-    "@whisper-cpp-node/win32-x64": "0.2.3"
+    "@whisper-cpp-node/darwin-arm64": "0.2.3",
+    "@whisper-cpp-node/win32-x64": "0.2.4"
   },
   "devDependencies": {
     "@types/node": "^20.0.0",