Offline speech processing / speech-to-text / text-to-speech
A cross-platform embedded/edge speech toolkit that runs ASR, TTS, VAD, diarization and more offline via ONNX Runtime with bindings in many languages.
Solves the same problem with a different architecture or at a different layer. Expect to rewrite the integration.
Use when you want actively maintained SOTA ASR/TTS models and training pipelines in Python, and can trade on-device/ONNX portability for server-side depth.
Easy-to-use Speech Toolkit including Self-Supervised Learning model, SOTA/Streaming ASR with punctuation, Streaming TTS with text frontend, Speaker Verification System, End-to-End Speech Translation and Keyword Spotting. Won NAACL2022 Best Demo Award.
Use when you need a research-oriented PyTorch toolkit for custom speech models and recipes, not a pre-packaged embedded runtime.
A PyTorch-based Speech Toolkit
Use when you need extremely low-latency ASR/TTS with a lightweight C++ engine for voice agents, and can accept fewer languages and missing diarization/audio-processing features.
Very low latency speech to text, intent recognition, and text to speech, for building voice agents and interfaces
Use when you are building an Apple-only app and want native CoreML-accelerated speech models instead of ONNX cross-platform.
Frontier CoreML audio models in your apps — text-to-speech, speech-to-text, voice activity detection, and speaker diarization. In Swift, powered by SOTA open source.
Use for Apple Silicon apps wanting MLX/CoreML-based ASR/TTS/VAD with tight Swift integration, rather than cross-platform ONNX.
AI speech toolkit for Apple Silicon — ASR, TTS, speech-to-speech, VAD, and diarization powered by MLX and CoreML
Use when you want a Python ASR toolkit with industrial-grade streaming ASR, VAD, punctuation, and OpenAI-compatible/MCP serving on servers, not embedded devices.
Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.
Use when you need a full generative-AI framework with trainable speech models and LLM integration, at the cost of heavy GPU/server requirements.
A scalable generative AI framework built for researchers and developers working on Large Language Models, Multimodal, and Speech AI (Automatic Speech Recognition and Text-to-Speech)
Use when you need a flexible end-to-end speech processing toolkit for research and training across many tasks, not a lightweight edge runtime.
End-to-End Speech Processing Toolkit
Use when you want a production-first end-to-end ASR training/inference toolkit with streaming support and your focus is server/cloud deployment.
Production First and Production Ready End-to-End Speech Recognition Toolkit
Use when you need an all-in-one simultaneous speech recognition/translation/synthesis model and can accept a research-oriented Python package.
StreamSpeech is an “All in One” seamless model for offline and simultaneous speech recognition, speech translation and speech synthesis.
Use for on-device speech AI on Apple Silicon with a Swift-native library, when you don't need cross-platform Android/Windows support.
On-device Speech AI for Apple Silicon
An earlier or more limited way to do the job. Still the right call when you need something small, proven, or CPU-only.
Use when you only need mature offline ASR with many language models and lightweight bindings, and can forego TTS/diarization/other tasks.
Offline speech recognition API for Android, iOS, Raspberry Pi and servers with Python, Java, C# and Node
Use when you prefer ncnn over ONNX Runtime for some embedded platforms and need only real-time ASR/VAD from the same k2-fsa family.
Real-time speech recognition and voice activity detection (VAD) using next-gen Kaldi with ncnn without Internet connection. Support iOS, Android, Linux, macOS, Windows, Raspberry Pi, VisionFive2, LicheePi4A etc.
Use when you need high-performance on-device multilingual TTS alone on Apple platforms and don't need ASR/VAD/diarization.
Lightning-Fast, On-Device, Multilingual TTS — running natively via ONNX.
Does the same job, but ships as an app. Useful to a person, not swappable into a codebase.
An API server/tool that wraps Whisper-style streaming ASR and diarization rather than an embeddable SDK.
Real-time, local speech-to-text with streaming ASR, speaker diarization, translation, and OpenAI/Deepgram-compatible APIs.
A webservice wrapper around OpenAI Whisper for REST-based transcription, not an embeddable speech toolkit.
OpenAI Whisper ASR Webservice API
A local dictation app that adds LLM post-processing, not a developer SDK.
Fully local, private and cross platform Speech-to-Text with LLM Post-processing
A consumer voice-input app that post-processes speech with LLMs, not a reusable speech library.
Hold a key, speak, release — AI-polished text appears at your cursor in any app. Open-source voice input for macOS & Windows. (按住快捷键说话，松开即得润色后的文字)
These projects were analysed and named sherpa-onnx among their alternatives. The relationship is not symmetric — how sherpa-onnx rates them is a separate judgement, made when sherpa-onnx is analysed in its own right.
calls sherpa-onnx “actively maintained offline STT with onnxruntime, broader features, and no Kaldi dependency.”
Offline speech recognition API for Android, iOS, Raspberry Pi and servers with Python, Java, C# and Node
calls sherpa-onnx “For a broader ONNX-based speech toolkit with TTS plus STT/VAD, if you also need other speech features.”
Lightning-Fast, On-Device, Multilingual TTS — running natively via ONNX.
calls sherpa-onnx “choose when you need a mature, lightweight, Apache-2.0 on-device speech toolkit with STT, TTS, VAD, diarization, and broader platform support.”
Very low latency speech to text, intent recognition, and text to speech, for building voice agents and interfaces
calls sherpa-onnx “Choose when you need fully offline TTS alongside ASR/VAD/diarization in an embedded-friendly onnxruntime package.”
Use Microsoft Edge's online text-to-speech service from Python WITHOUT needing Microsoft Edge or Windows or an API key
calls sherpa-onnx “Choose when you need lightweight offline inference on embedded/edge devices with ONNX models.”
A PyTorch-based Speech Toolkit
calls sherpa-onnx “when deploying on embedded/CPU targets with C++ and ONNX runtime instead of a Python/PyTorch server”
Real-time, local speech-to-text with streaming ASR, speaker diarization, translation, and OpenAI/Deepgram-compatible APIs.
calls sherpa-onnx “Choose when you need offline on-device speech tasks with ONNX runtime for embedded systems, as a lightweight inference-only alternative.”
End-to-End Speech Processing Toolkit
calls sherpa-onnx “choose if you need a cross-platform C++/embedded inference engine that includes TTS.”
A TTS that fits in your CPU (and pocket)
calls sherpa-onnx “choose for embedded/on-device TTS via ONNX; not a TTS-only service.”
EmotiVoice 😊: a Multi-Voice and Prompt-Controlled TTS Engine
calls sherpa-onnx “Choose when you need cross-platform on-device speech processing with onnxruntime support and no Apple requirement.”
A text-to-speech (TTS), speech-to-text (STT) and speech-to-speech (STS) library built on Apple's MLX framework, providing efficient speech analysis on Apple Silicon.
calls sherpa-onnx “Embedded-friendly runtime with TTS and ASR via ONNX; use it for offline/low-resource deployment without internet.”
High-quality multi-lingual text-to-speech library by MyShell.ai. Support English, Spanish, French, Chinese, Japanese and Korean.