Whispering Tiger

Depending on the selected models, Whispering Tiger can understand or translate between 100 to 200+ languages. Coverage varies per model family (see sections below). Many pipelines allow language autodetection.

• ASR (Speech-to-Text): Typical Whisper-family models cover ~99 languages; with support to translate to English. Other models already support translation directly to other languages as well.
• Text Translation: Multi-lingual models (e.g. NLLB-200, M2M100) reach 200 languages; With local Large-Language Models (LLMs) with high translation accuracy.
• TTS: Many Engines; some offer dozens of voices across 100+ languages, others focus on a single language. With some even supporting voice cloning with a short audio sample.
• OCR: OCR backends (e.g. Easy OCR, GOT OCR 2.0, Phi-4 or Tesseract via plugin) support large multilingual sets.

• Whisper original (multiple sizes: tiny → large-v3)
• Transformer-Whisper (Supporting Flash-Attention)
• Faster-Whisper / CTranslate2 optimized variants
• With support for custom user-provided models
• Seamless M4T
• MMS
• Speech T5
• Wav2Vec Bert 2.0
• NeMo Canary
• Phi-4 (LLM)
• Voxtral (LLM)
• + Cloud Models like OpenAI, Google, etc. via Plugins

UI lets you pick size vs accuracy, precision / quantization, used hardware device and VAD (voice activity detection) options.

• Real-time streaming or batch file transcription
• Automatic language detection
• Timestamped segments with optional word-level timing (when backend supports it)
• Customizable VAD sensitivity and chunk parameters
• Segment post-processing (casing, punctuation, number formatting)
• On-the-fly translation (transcribe + translate to target)
• Profile-based hardware tuning (threads, GPU, fp16/int8/4-bit quantization)
• Subtitle generation and translation (SRT, VTT, etc.)

• NLLB-200 + CTranslate2 optimized with (200 languages)
• M2M100 (100 languages)
• Seamless M4T (101 languages)
• Phi-4 (23 languages)
• Voxtral (13 languages)
• + Cloud Models like DeepL, OpenAI, Google, etc. via Plugins

• Translation of text files or clipboard contents
• Real-time speech → text → translated text workflow
• Chained pipelines: ASR → MT → TTS for spoken output
• Automatic source language detection & fallback strategies

• Silero (fast, CPU-friendly, multilingual)
• Kokoro (fast, high-quality, multilingual)
• F5/E2 (high-quality, expressive TTS with voice-cloning)
• Zonos (high-quality, expressive TTS with voice-cloning and many languages)
• Orpheus (high-quality, expressive TTS with voice-cloning)
• + Plugins for more TTS options

  • ---

    Bark with expressive generation
  • Coqui TTS (neural multi-speaker models)
  • Voice Conversion / Cloning plugins (RVCv2)
  • Cloud Models like OpenAI, Google etc.

• Selectable voice, speed, pitch (engine dependent)
• Text normalization & sentence splitting
• Streamed audio generation for faster playback
• Audio output device selection for multiple devices without the need of VoiceMeeter
• For output into voice-chat applications like Games, a virtual audio device is required

• Easy OCR (fast & accurate for many scripts)
• GOT OCR 2.0 (accurate for many scripts and complicated layouts)
• Phi-4 (accurate for many scripts and complicated layouts with Large-Language-Model)
• Tesseract (if plugin installed)
• Game / screen region capture overlays

• Configurable screen area polling
• Automatic text translation of OCR results
• Overlay and subtitle-mode output
• Text-to-Speech playback of translated lines
• Filtering & deduplication of repeated frames
• Logged transcripts for later review

Extend core capabilities without modifying the base application.

• Install / update / remove inside the UI
• Easy Plugin API (see plugin documentation)
• Extensive Plugin Configuration UI
• Events (stt, tts, sts and further custom plugin events like on_plugin_tts, on_audio_processor etc.)
• Access to application, model & pipeline configuration
• Fast Audio processing Functions usable by Plugins like resampling, normalization, etc.
• Community-driven enhancements (see plugin list)

• Profiles for different tasks (gaming, meetings, game translation etc.)
• Custom hotkeys (start/stop, push-to-talk)
• Auto model download
• Update management for installed plugins

• GPU acceleration (CUDA, CPU, with possible ROCm support in the future)
• Mixed precision / fp16 inference
• Quantized model support (int8 / 4-bit) for low VRAM systems
• Thread & batch tuning per model

• On-demand model loading
• Disk cache for reused weights
• Offline mode as main focus with optional Plugins for online functionality

• Tabbed UI for distinct tasks (ASR, Translation, TTS, OCR)
• Central YAML Configuration file for easy sharing & collaboration (make sure to exclude sensitive information when sharing)
• Dark theme focused on readability
• Status indicators
• Websocket support for integration into other applications like Streaming software or Virtual Reality overlays
• VRChat Support via OSC

• Subtitle formats (SRT / VTT) (when plugin enabled)
• Plain text transcripts
• Timestamp annotation
• HTML files with integration for overlays and other use cases

• All processing can run fully offline
• No audio/text sent to external servers unless a plugin explicitly does so (clearly indicated)
• User-controlled model and cache directories
• Transparent logs; easy to audit behavior
• Optional Error Reporting to help improve the system with user selective attached information

No. New plugins, models and features appear frequently. For the freshest additions check the GitHub repositories and plugin index.

Yes. Model size, hardware (CPU/GPU), quantization, and concurrent tasks all influence throughput and latency. Profiles help tune these.

Open an issue on the main UI repository on GitHub, Join the Discord server
or send a report using the in-app feedback tool.