sherpa-onnx public API documentation

This documentation covers the public native APIs shipped in:

• c-api.h — the C API
• cxx-api.h — the C++ wrapper built on top of the C API

These headers expose the main sherpa-onnx inference features for native applications and for language bindings that need a stable ABI.

What is documented here

The generated docs include the public APIs for:

• streaming ASR
• non-streaming ASR
• keyword spotting
• voice activity detection
• offline text-to-speech
• spoken language identification
• speaker embedding extraction and speaker management
• audio tagging
• offline and online punctuation
• linear resampling
• offline speaker diarization
• offline and online speech enhancement

Model-specific documentation

Each model family has its own documentation page with config examples:

• Non-Streaming (Offline) ASR Models — Non-streaming ASR: Zipformer Transducer, Zipformer CTC, Whisper, SenseVoice, Paraformer, Moonshine, FireRedAsr, Dolphin, Canary, Cohere, WeNet, Omnilingual, FunASR Nano, Qwen3, MedASR, TeleSpeech, GigaAM v2, Parakeet TDT, NeMo CTC
• Streaming (Online) ASR Models — Streaming ASR: Transducer (Zipformer), Nemotron, Paraformer, Zipformer2 CTC, T-One CTC
• Text-to-Speech (TTS) Models — Text-to-Speech: Kokoro, VITS (Piper), Matcha, Kitten, ZipVoice, Pocket, Supertonic
• Voice Activity Detection (VAD) — Voice Activity Detection: Silero VAD, Ten VAD
• Audio Tagging — Audio Tagging: Zipformer, CED
• Punctuation Restoration — Punctuation: Offline (CT-Transformer), Online (CNN-BiLSTM)
• Speech Enhancement / Denoising — Speech Enhancement: GTCRN, DPDFNet (offline and online)
• Source Separation — Source Separation: Spleeter, UVR
• Offline Speaker Diarization — Speaker Diarization: Pyannote segmentation + embedding clustering
• Speaker Embedding Extraction and Management — Speaker Embedding: extraction, enrollment, search, verification
• Spoken Language Identification — Spoken Language Identification: Whisper-based
• Keyword Spotting — Keyword Spotting: Zipformer KWS
• Linear Resampler — Linear Resampler

The C API also includes HarmonyOS-specific constructor variants where applicable.

Which header should I use?

Use c-api.h if you are:

• writing C code
• building FFI bindings for other languages
• integrating through a plain C ABI

Use cxx-api.h if you are:

• writing C++ code directly
• preferring RAII wrappers over manual destroy/free calls
• preferring std::string, std::vector, and move-only wrapper classes

Common ownership rules

For the C API:

• objects created by SherpaOnnxCreate*() are usually destroyed with a matching SherpaOnnxDestroy*()
• result snapshots, returned strings, and returned arrays must be released with the specific matching free/destroy function documented on each API
• some helpers return pointers to statically owned strings; those must not be freed

For the C++ API:

• wrapper classes are move-only and use RAII
• copied result objects are returned as standard C++ value types
• callers normally do not need to manage the underlying C pointers directly

Typical workflow

For both APIs, the usual flow is:

1. create and fill a config object
2. create the engine or recognizer
3. create a stream if the feature is stream-based
4. feed audio or text
5. run decode/compute/generate
6. read back results
7. destroy resources, or let the C++ wrappers clean them up automatically

Recommended entry points

Start with:

• c-api.h for the plain C API
• cxx-api.h for the C++ wrapper

Representative example programs live in:

• c-api-examples/
• cxx-api-examples/

Useful examples include:

Offline ASR (C API):

• whisper-c-api.c
• sense-voice-c-api.c
• paraformer-c-api.c
• moonshine-c-api.c
• zipformer-c-api.c
• nemo-parakeet-c-api.c
• nemo-giga-am-v2-c-api.c
• nemo-ctc-c-api.c
• cohere-transcribe-c-api.c
• dolphin-ctc-c-api.c

Streaming ASR (C API):

• streaming-zipformer-c-api.c
• streaming-nemotron-c-api.c
• streaming-paraformer-c-api.c

TTS (C API):

• kokoro-tts-en-c-api.c
• pocket-tts-en-c-api.c
• offline-tts-c-api.c

Other features (C API):

• vad-whisper-c-api.c
• audio-tagging-c-api.c
• speaker-identification-c-api.c
• offline-speaker-diarization-c-api.c
• spoken-language-identification-c-api.c
• kws-c-api.c

C++ API examples:

• streaming-zipformer-cxx-api.cc
• streaming-nemotron-cxx-api.cc
• paraformer-cxx-api.cc
• nemo-ctc-cxx-api.cc
• offline-tts-piper-cxx-api.cc
• pocket-tts-en-cxx-api.cc
• vad-cxx-api.cc
• offline-speaker-diarization-cxx-api.cc
• speaker-identification-cxx-api.cc
• spoken-language-identification-cxx-api.cc

Generating the documentation

From sherpa-onnx/c-api/, run:

doxygen Doxyfile

HTML output is written to:

doxygen-docs/html/