Dolphin CN-Dialect Base · OpenASR

Chinese multi-dialect speech recognition, base tier -- a compact 140M WeNet E-Branchformer (CTC + attention) for Sichuan and 22 regional dialects

Native speech-to-text in the OpenASR runtime — engineered for peak performance on CPU & GPU, no Python at inference time.

---

✨ Highlights

• 🀄 22 Chinese dialects, base tier — the same WeNet E-Branchformer dialect coverage as Dolphin CN-Dialect Small (Sichuan/川话, Wu, Cantonese, Minnan, Shanghainese and more), at a fraction of the size
• 🪶 140M parameters — roughly a third the width of the small.cn checkpoint (512 vs 768 d_model, 6 vs 12 layers), for tighter RAM and faster CPU decode when the small tier is overkill
• 🧩 Joint CTC + attention — the same E-Branchformer encoder + Transformer decoder recipe with CTC/attention rescoring, verified against a shape-derived runtime contract shared with the rest of the Dolphin family
• 🐬 Chinese-focused, mixed char/BPE vocab — a character vocabulary for Chinese with SentencePiece word-piece tokens for code-switched English, purpose-built for zh audio including heavy accents
• 🦀 Native in OpenASR — .oasr packs run with no Python at inference, engineered for peak performance on CPU & GPU

🚀 Quickstart

# 1. Install the OpenASR CLI  ·  https://openasr.org
# 2. Pull a build (pick a quant — see the table below)
openasr pull dolphin-cn-dialect-base:fp16

# 3. Transcribe
openasr transcribe audio.wav --model dolphin-cn-dialect-base

All builds for this model:

openasr pull dolphin-cn-dialect-base:fp16
openasr pull dolphin-cn-dialect-base:q8
openasr pull dolphin-cn-dialect-base:q4

📦 Available builds

_{RTF = real-time factor on the shared 11s JFK clip (out-of-distribution English, drift signal only) plus an in-language Mandarin sanity clip (lower is faster); RAM peak measured per pack in an isolated subprocess. ΔWER compares each quantized build's JFK + zh sanity clip transcript to this model's fp16 JFK + zh sanity clip transcript, so it measures quantization drift rather than absolute recognition accuracy. fp16 is the recommended default — near-reference quality at a fraction of the footprint.}

🧠 About Dolphin CN-Dialect Base

Dolphin CN-Dialect Base is the 140M "base" tier of DataoceanAI's Chinese multi-dialect speech-recognition line, built on the same Dolphin / WeNet recipe as the larger Dolphin CN-Dialect Small: an E-Branchformer encoder + Transformer decoder trained with a joint CTC + attention objective over a mixed character/BPE vocabulary. It covers the same Sichuan (川话)-forward set of 22 Chinese dialects (Wu, Cantonese, Minnan, Shanghainese and more) as its small.cn sibling, but at roughly a third of the encoder/decoder width (512 vs 768 d_model, 6 vs 12 layers) -- a smaller RAM/CPU footprint for deployments where the small tier's accuracy headroom is not needed. Unlike small.cn, this base.cn checkpoint does not ship a trained hotword deep-biasing module. This OpenASR repo repackages the weights as .oasr packs that run natively in the OpenASR runtime -- no Python at inference, all decoding local. It ships in fp16 (maximum fidelity, recommended), q8_0, and q4_k builds.

Note: this model does not emit punctuation. Its upstream training corpus is transcribed without punctuation marks, so the decoder never predicts a punctuation token -- there is no setting to enable it. Transcripts are plain, unpunctuated text by design.

⚙️ How these packs were made

Converted from DataoceanAI/dolphin-cn-dialect-base with the OpenASR importer:

openasr model-pack import dolphin <src> <out>.oasr \
  --package-id dolphin-cn-dialect-base --quantization {fp16,q8-0,q4-k}

The .oasr container is GGUF-backed; packs use zero-copy mmap weight binding and graph buffer reuse to keep peak memory low.

⚖️ License

These packs inherit the upstream model's license: Apache-2.0 (source). OpenASR packaging retains the upstream copyright and NOTICE; the only modifications are format conversion and quantization.

🙏 Acknowledgements

This pack is a redistribution of Dolphin CN-Dialect Base (base.cn), created and open-sourced by DataoceanAI (DataoceanAI/dolphin-cn-dialect-base). All credit for the original architecture, training, and weights belongs to the authors; the license is inherited from and identical to the upstream model (Apache-2.0). The model builds on the Dolphin multilingual ASR project and the WeNet E-Branchformer / joint CTC-attention recipe -- thank you to the Dolphin and WeNet teams and to DataoceanAI for releasing their work openly. OpenASR only performs format conversion, quantization, runtime verification, and local-inference adaptation.

🔗 Links

• 🦀 OpenASR — https://github.com/QuintinShaw/openasr
• 🌐 Website — https://openasr.org
• 🤗 Upstream model — DataoceanAI/dolphin-cn-dialect-base