Third-party software and models
Private beta document · last updated October 24, 2026. Plain language on purpose — if anything here is unclear, ask us before you rely on it.
Vearo's Android app runs its speech models on your phone. Voice activity detection, transcription, speaker diarization, speech enhancement and the candidate SenseVoiceSmall model all run on your device inside the Vearo app, using the open-source software and models listed below. Nothing on this page runs on Vearo's servers; the AI-provider analysis that does is described in the Privacy Policy.
This page is the attribution summary. The full license texts ship inside the app under Settings › About › Open-source licenses, and a copy is available on request through the contact address in your invitation during the private beta. Model files are bundled into the app from the upstream releases listed below.
Software
- whisper.cpp (incl. ggml) — on-device speech-to-text engine, pinned at v1.8.6. Upstream: github.com/ggml-org/whisper.cpp. License: MIT — Copyright (c) 2023-2026 The ggml authors.
- sherpa-onnx 1.13.3 — ONNX inference runtime for voice activity detection, speaker diarization, speech enhancement and SenseVoice. Upstream: github.com/k2-fsa/sherpa-onnx. License: Apache License 2.0 (the upstream LICENSE file carries no copyright line; project: k2-fsa / Xiaomi Corporation).
- ONNX Runtime — inference engine bundled inside the sherpa-onnx Android library. Upstream: github.com/microsoft/onnxruntime. License: MIT — Copyright (c) Microsoft Corporation.
Models
- Whisper small.en model (ggml-small.en-q5_1.bin) — default English transcription model: OpenAI Whisper weights in ggml format, via the conversions published at huggingface.co/ggerganov/whisper.cpp (q5_1 quantization by the whisper.cpp project). Upstream: github.com/openai/whisper. License: MIT — Copyright (c) 2022 OpenAI.
- SenseVoiceSmall (sense-voice-small.int8.onnx, sense-voice-small-tokens.txt) — candidate speech, emotion and audio-event model by FunAudioLLM / FunASR. The bytes are sherpa-onnx's int8 conversion of ModelScope iic/SenseVoiceSmall, release asset sherpa-onnx-sense-voice-zh-en-ja-ko-yue-int8-2024-07-17. Upstream: github.com/FunAudioLLM/SenseVoice · modelscope.cn/models/iic/SenseVoiceSmall. License (model weights): FunASR Model Open Source License Agreement, Version 1.1 — Copyright Alibaba Group. The model is used, copied and shared under that agreement with this attribution, and the model name SenseVoiceSmall is retained verbatim. SenseVoice source code: MIT — Copyright (c) 2025 FunASR.
- Silero VAD v5 (silero_vad_v5.onnx) — voice activity detection. Upstream: github.com/snakers4/silero-vad. License: MIT — Copyright (c) 2020-present Silero Team.
- CAM++ speaker embedding (3dspeaker_speech_campplus_sv_zh_en_16k-common_advanced.onnx) — speaker voiceprints for diarization and the owner voice profile; model speech_campplus_sv_zh_en_16k-common_advanced (ModelScope iic, 3D-Speaker project). Upstream: github.com/modelscope/3D-Speaker · modelscope.cn/models/iic/speech_campplus_sv_zh_en_16k-common_advanced. License: Apache License 2.0 — Copyright 3D-Speaker contributors, Alibaba DAMO Academy (ModelScope iic).
- pyannote segmentation 3.0 (pyannote_segmentation_v3_int8.onnx) — speaker-change detection used to sub-split turns; int8 ONNX conversion distributed by sherpa-onnx. Upstream: huggingface.co/pyannote/segmentation-3.0. License: MIT (model card license: mit) — Copyright (c) 2020 CNRS (pyannote.audio).
- GTCRN speech enhancement (gtcrn_simple.onnx) — light noise suppression before detection and transcription; ONNX export distributed by sherpa-onnx. Upstream: github.com/Xiaobin-Rong/gtcrn. License: MIT — Copyright (c) 2024 Rong Xiaobin.
Keeping this page accurate
Each license above was read at the linked upstream on October 24, 2026. If a component or model changes in the app, this page and the in-app list are updated together. Questions about a notice can go to the contact address in your invitation during the private beta.