OpenASR/dolphin-base

Dolphin Base · OpenASR

Multilingual speech recognition across 40 languages, base tier -- a compact 140M WeNet/ESPnet E-Branchformer (CTC + attention)

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

---

✨ Highlights

• 🌏 40 languages, base tier — the same multilingual E-Branchformer coverage as Dolphin Small (South Asian, Southeast Asian, Central Asian/Turkic, Chinese/Cantonese), at a fraction of the size
• 🪶 140M parameters — roughly a third the width of the small checkpoint (512 vs 768 d_model, fewer 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
• 🐬 SentencePiece BPE vocab — a shared subword vocabulary across all 40 languages (distinct from the cn-dialect family's fixed character vocab)
• 🦀 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-base:fp16

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

All builds for this model:

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

📦 Available builds

Quant	File (.oasr)	Size	RAM peak	RTF · M1 CPU	RTF · M1 GPU	ΔCER vs fp16
fp16	dolphin-base-fp16.oasr	287 MB	1.92 GB	0.15×	0.14×	0.0%
q8_0	dolphin-base-q8_0.oasr	158 MB	1.76 GB	0.15×	0.16×	0.0%
q4_k	dolphin-base-q4_k.oasr	90 MB	1.70 GB	0.13×	0.13×	8.8%

_{RTF = real-time factor on the shared 11s JFK clip (out-of-distribution, drift signal only) plus an in-language Mandarin sanity clip (lower is faster); RAM peak measured per pack in an isolated subprocess. ΔCER 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 Base

Dolphin Base is the 140M "base" tier of DataoceanAI's multilingual Dolphin speech- recognition line, built on the same Dolphin / ESPnet recipe as the larger Dolphin Small: an E-Branchformer encoder + Transformer decoder trained with a joint CTC + attention objective over a shared SentencePiece BPE vocabulary spanning the card's advertised 40 languages (South Asian, Southeast Asian, Central Asian/Turkic, and Chinese including Cantonese as yue), at roughly a third of the small tier's encoder/decoder width -- a smaller RAM/CPU footprint for deployments where the small tier's accuracy headroom is not needed. Like dolphin-small, this checkpoint collapses this product's own Chinese-dialect granularity into a single zh (the dedicated dolphin-cn-dialect-small/-base packs cover per-dialect prompting). 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.

Verification status: this pack is staged in a private repo, not yet publicly listed. Local verification so far covers Mandarin (zh) sanity-checked against the upstream architecture and bit-stable at fp16/q8_0, with a small (~9% CER) drift at q4_k versus fp16 on the sanity clip; Japanese (ja), one of the 40 advertised languages, has not yet had a native-speaker listening review and must get one before this model is made public.

⚙️ How these packs were made

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

openasr model-pack import dolphin <src> <out>.oasr \
  --package-id dolphin-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 Base, created and open-sourced by DataoceanAI (DataoceanAI/dolphin-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 ESPnet E-Branchformer / joint CTC-attention recipe -- thank you to the Dolphin and ESPnet 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-base