Model card: custom vocabulary, speaker labels and updated limitations

Adds the custom vocabulary and the speaker-label example (with NVIDIA's Nemotron-3-Diarization) to How to use, names GPT-Transcribe as on par in the results, updates the limitations to match, adds a link to the demo Space, and follows the inference code's changes (Python 3.11+, example file names).

README.md

@@ -22,6 +22,7 @@ tags:
 # Haas-ja-v1.0
 
 [![Code: GitHub](https://img.shields.io/badge/Code-GitHub-black?logo=github)](https://github.com/third-intelligence/Haas-ASR)
+[![Demo: Hugging Face Space](https://img.shields.io/badge/Demo-Hugging%20Face%20Space-yellow?logo=huggingface)](https://huggingface.co/spaces/third-intelligence/Haas-ja-v1.0)
 [![Blog](https://img.shields.io/badge/Blog-Third%20Intelligence-green)](TODO-blog-url)
 [![Benchmark: asr-ja-bench](https://img.shields.io/badge/Benchmark-asr--ja--bench-orange)](https://huggingface.co/datasets/third-intelligence/asr-ja-bench)
 [![Metrics: jacer](https://img.shields.io/badge/Metrics-jacer-purple)](https://github.com/third-intelligence/jacer)
@@ -67,7 +68,7 @@ The full architecture configuration is in [`config.yaml`](config.yaml).
 
 ## How to use
 
-Requirements: Linux with an NVIDIA GPU, Python 3.10+ and NeMo 2.7. The prediction
+Requirements: Linux with an NVIDIA GPU, Python 3.11+ and NeMo 2.7. The prediction
 network runs on kernels from
 [flash-linear-attention](https://github.com/fla-org/flash-linear-attention), which
 needs Triton and CUDA, so CPU and macOS are not supported.
@@ -82,7 +83,7 @@ dependency versions we tested:
 git clone https://github.com/third-intelligence/Haas-ASR.git
 cd Haas-ASR
 uv sync
-uv run python examples/quickstart.py --model third-intelligence/Haas-ja-v1.0 audio.wav
+uv run python examples/quickstart.py --model third-intelligence/Haas-ja-v1.0 example.wav
 ```
 
 **Use it in your own project**
@@ -99,7 +100,7 @@ Then transcribe:
 import haas_asr
 
 model = haas_asr.load_model("third-intelligence/Haas-ja-v1.0")
-audio = haas_asr.load_audio("audio.wav")
+audio = haas_asr.load_audio("example.wav")
 output = model.transcribe([audio])
 print(output[0].text)
 ```
@@ -112,10 +113,42 @@ For recordings longer than 30 seconds, use `transcribe_longform`. It splits the
 at quiet points (detected by volume) and transcribes the pieces in batches:
 
 ```python
-audio = haas_asr.load_audio("lecture.flac")
+audio = haas_asr.load_audio("example.flac")
 text = haas_asr.transcribe_longform(model, audio, batch_size=64)
 ```
 
+**Custom vocabulary**
+
+Names and terms the model rarely heard in training can be listed, so that it
+recognizes them more often and writes them as you want:
+
+```python
+vocabulary = haas_asr.Vocabulary(
+    {"Haas": ["ハース"], "Third Intelligence": ["サードインテリジェンス"]}
+)
+texts = haas_asr.transcribe(model, [audio], vocabulary=vocabulary)  # short clips
+text = haas_asr.transcribe_longform(model, audio, vocabulary=vocabulary)
+```
+
+Each word maps its written form to its readings, the ways Haas would write it (usually
+in kana; several, or none if Haas already writes it as listed). The readings are
+boosted while decoding with NeMo's
+[GPU phrase boosting](https://docs.nvidia.com/nemo-framework/user-guide/latest/nemotoolkit/asr/asr_customization/word_boosting.html),
+and replaced by the written form wherever found. Boosting can also change text where
+no listed word was said, so audio without speech is not boosted, and each audio is
+also decoded without the words: a change is kept only where it includes a whole listed
+word. Both decodes share one encoder pass, so a vocabulary costs about 1.3 times the
+time. See the
+[repository](https://github.com/third-intelligence/Haas-ASR#custom-vocabulary) for
+details.
+
+**Speaker labels**
+
+Combined with NVIDIA's speaker diarization model
+[Nemotron-3-Diarization](https://huggingface.co/nvidia/Nemotron-3-Diarization), Haas
+produces transcripts with speaker labels. The repository has a
+[runnable example](https://github.com/third-intelligence/Haas-ASR#speaker-labels-with-nemotron-3-diarization).
+
 ## Training data
 
 The model was trained on about 46,000 hours of audio from public corpora
@@ -173,8 +206,8 @@ In each cell, the top value is CER strict and the bottom values are CER flex / k
 (%, lower is better). Models are sorted by weighted CER strict.
 
 Haas-ja-v1.0 reaches a weighted CER strict of 6.12%, the best of the open models
-evaluated. It is competitive with commercial APIs, though MAI-Transcribe 2, Scribe v2 and
-Gemini 3.1 Pro remain more accurate.
+evaluated. It is competitive with commercial APIs: on par with GPT-Transcribe (6.07%),
+while MAI-Transcribe 2, Scribe v2 and Gemini 3.1 Pro remain more accurate.
 
 ![CER of each model on the five datasets and RTFx](assets/cer_table.png)
 
@@ -210,13 +243,14 @@ running a service. API costs are what OpenRouter billed during the evaluation.
   utterance at once. It can still produce live captions by re-transcribing the current
   phrase as audio arrives (see the demo in the repository), but low-latency streaming
   may need further tuning.
-- **Non-verbal information.** The model transcribes what is said, not who says it or how:
-  speaker identity, emotion and other non-verbal information are not recognized.
-- **Custom vocabulary.** There is no built-in way to supply your own domain terms or
-  proper nouns.
+- **Non-verbal information.** On its own, the model transcribes what is said, not who
+  says it or how: speaker identity, emotion and other non-verbal information are not
+  recognized. Speaker labels are possible by combining it with a diarization model (see
+  How to use).
 - **Output format.** The output style cannot be controlled, for example whether fillers
   are included. The model tends to omit fillers, and its choices of kanji, kana and
-  numerals follow its training data.
+  numerals follow its training data. The custom vocabulary can fix the spelling of
+  individual words, but not the style as a whole.
 
 ## Acknowledgements
 
@@ -228,8 +262,10 @@ It builds on [NVIDIA NeMo](https://github.com/NVIDIA/NeMo) and the
 prediction network uses Kimi Delta Attention from
 [Kimi Linear](https://arxiv.org/abs/2510.26692), run on the kernels of
 [flash-linear-attention](https://github.com/fla-org/flash-linear-attention), and it was
-trained with [Lhotse](https://github.com/lhotse-speech/lhotse). We thank the authors
-of these projects for making them openly available.
+trained with [Lhotse](https://github.com/lhotse-speech/lhotse). The speaker-label
+example uses NVIDIA's
+[Nemotron-3-Diarization](https://huggingface.co/nvidia/Nemotron-3-Diarization)
+(OpenMDW-1.1). We thank the authors of these projects for making them openly available.
 
 We also thank the creators of [ReazonSpeech](https://huggingface.co/datasets/reazon-research/reazonspeech),
 [Common Voice](https://commonvoice.mozilla.org) and