Refactored all ASR collections documentation #15542
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,339 @@ | ||
| .. _asr-checkpoints-list: | ||
|
|
||
| ======================= | ||
| ASR Model Checkpoints | ||
| ======================= | ||
|
|
||
| This page lists all supported ASR model checkpoints released by NVIDIA NeMo. | ||
| Benchmark scores for each model can be found on its `HuggingFace model card <https://huggingface.co/nvidia>`__. | ||
|
|
||
| Glossary | ||
| -------- | ||
|
|
||
| .. list-table:: | ||
| :header-rows: 1 | ||
|
|
||
| * - Term | ||
| - Definition | ||
| * - **ASR** | ||
| - Automatic Speech Recognition — transcribing speech to text | ||
| * - **AST** | ||
| - Automatic Speech Translation — translating speech to text from one language to another | ||
| * - **AED** | ||
| - Attention Encoder-Decoder — autoregressive decoder using cross-attention (Canary family) | ||
| * - **CTC** | ||
| - Connectionist Temporal Classification — non-autoregressive decoder | ||
| * - **RNN-T** | ||
| - Recurrent Neural Network Transducer — autoregressive streaming-friendly decoder | ||
| * - **TDT** | ||
| - Token-and-Duration Transducer — extends RNN-T with duration prediction for faster inference | ||
| * - **Hybrid** | ||
| - Joint RNN-T + CTC model — both decoders trained together, either usable at inference | ||
| * - **PnC** | ||
| - Punctuation and Capitalization in the output | ||
| * - **Streaming** | ||
| - Real-time / cache-aware inference capability | ||
| * - **EU4** | ||
| - Multilingual: English, German, Spanish, French | ||
| * - **EU25** | ||
| - Multilingual: 25 European languages (de, en, es, fr, it, pl, pt, nl, ru, uk, be, hr, cs, bg, da, et, fi, el, hu, lv, lt, mt, ro, sk, sl, sv) | ||
|
|
||
|
|
||
| Canary Models (AED) | ||
| ------------------- | ||
|
|
||
| Multi-task encoder-decoder models supporting ASR, AST, PnC, and timestamps across multiple languages. | ||
|
|
||
| .. list-table:: | ||
| :header-rows: 1 | ||
|
|
||
| * - Model | ||
| - Decoder | ||
| - Capabilities | ||
| - Languages | ||
| * - `canary-1b-v2 <https://huggingface.co/nvidia/canary-1b-v2>`__ | ||
| - AED | ||
| - ASR, AST, PnC, timestamps | ||
| - EU25 | ||
| * - `canary-qwen-2.5b <https://huggingface.co/nvidia/canary-qwen-2.5b>`__ | ||
| - AED | ||
| - ASR, AST, PnC, timestamps | ||
| - EU25 | ||
| * - `canary-1b-flash <https://huggingface.co/nvidia/canary-1b-flash>`__ | ||
| - AED | ||
| - ASR, AST, PnC, timestamps, fast | ||
| - EU4 | ||
| * - `canary-180m-flash <https://huggingface.co/nvidia/canary-180m-flash>`__ | ||
| - AED | ||
| - ASR, AST, PnC, timestamps, fast | ||
| - EU4 | ||
| * - `canary-1b <https://huggingface.co/nvidia/canary-1b>`__ | ||
| - AED | ||
| - ASR, AST, PnC | ||
| - EU4 | ||
|
|
||
|
|
||
| Parakeet Models (English) | ||
| -------------------------- | ||
|
|
||
| High-accuracy English ASR models with FastConformer encoder. | ||
|
|
||
| .. list-table:: | ||
| :header-rows: 1 | ||
|
|
||
| * - Model | ||
| - Decoder | ||
| - Capabilities | ||
| - Size | ||
| * - `parakeet-tdt-0.6b-v3 <https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3>`__ | ||
| - TDT | ||
| - ASR, PnC, timestamps | ||
| - 0.6B | ||
| * - `parakeet-tdt-0.6b-v2 <https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2>`__ | ||
| - TDT | ||
| - ASR, PnC, timestamps | ||
| - 0.6B | ||
| * - `parakeet-tdt-1.1b <https://huggingface.co/nvidia/parakeet-tdt-1.1b>`__ | ||
| - TDT | ||
| - ASR, timestamps | ||
| - 1.1B | ||
| * - `parakeet-tdt_ctc-1.1b <https://huggingface.co/nvidia/parakeet-tdt_ctc-1.1b>`__ | ||
| - Hybrid TDT+CTC | ||
| - ASR, timestamps | ||
| - 1.1B | ||
| * - `parakeet-tdt_ctc-0.6b-ja <https://huggingface.co/nvidia/parakeet-tdt_ctc-0.6b-ja>`__ | ||
| - Hybrid TDT+CTC | ||
| - ASR, timestamps, Japanese | ||
| - 0.6B | ||
| * - `parakeet-tdt_ctc-110m <https://huggingface.co/nvidia/parakeet-tdt_ctc-110m>`__ | ||
| - Hybrid TDT+CTC | ||
| - ASR, timestamps | ||
| - 110M | ||
| * - `parakeet-rnnt-1.1b <https://huggingface.co/nvidia/parakeet-rnnt-1.1b>`__ | ||
| - RNN-T | ||
| - ASR, timestamps | ||
| - 1.1B | ||
| * - `parakeet-rnnt-0.6b <https://huggingface.co/nvidia/parakeet-rnnt-0.6b>`__ | ||
| - RNN-T | ||
| - ASR, timestamps | ||
| - 0.6B | ||
| * - `parakeet-ctc-1.1b <https://huggingface.co/nvidia/parakeet-ctc-1.1b>`__ | ||
| - CTC | ||
| - ASR | ||
| - 1.1B | ||
| * - `parakeet-ctc-0.6b <https://huggingface.co/nvidia/parakeet-ctc-0.6b>`__ | ||
| - CTC | ||
| - ASR | ||
| - 0.6B | ||
| * - `parakeet-rnnt-110m-da-dk <https://huggingface.co/nvidia/parakeet-rnnt-110m-da-dk>`__ | ||
| - RNN-T | ||
| - ASR, Danish | ||
| - 110M | ||
|
|
||
|
|
||
| Streaming Models | ||
| ----------------- | ||
|
|
||
| Cache-aware models for real-time / low-latency inference. | ||
|
|
||
| .. list-table:: | ||
| :header-rows: 1 | ||
|
|
||
| * - Model | ||
| - Decoder | ||
| - Capabilities | ||
| - Languages | ||
| * - `nemotron-speech-streaming-en-0.6b <https://huggingface.co/nvidia/nemotron-speech-streaming-en-0.6b>`__ | ||
| - Hybrid | ||
| - ASR, streaming | ||
| - en | ||
|
Collaborator
There was a problem hiding this comment. Either extend the ISO code into full language name, or add another glossary at the end - we have to assume less technical people will be reading this too.
Collaborator
There was a problem hiding this comment. It may be more economical to just list the architecture and configure a list of supported language models, or maybe a matrix? |
||
| * - `multitalker-parakeet-streaming-0.6b-v1 <https://huggingface.co/nvidia/multitalker-parakeet-streaming-0.6b-v1>`__ | ||
| - RNN-T | ||
| - ASR, multitalker, streaming | ||
| - en | ||
| * - `parakeet_realtime_eou_120m-v1 <https://huggingface.co/nvidia/parakeet_realtime_eou_120m-v1>`__ | ||
| - RNN-T | ||
| - ASR, end-of-utterance, streaming | ||
| - en | ||
| * - `stt_en_fastconformer_hybrid_large_streaming_multi <https://huggingface.co/nvidia/stt_en_fastconformer_hybrid_large_streaming_multi>`__ | ||
| - Hybrid | ||
| - ASR, streaming, multiple look-aheads | ||
| - en | ||
| * - `stt_en_fastconformer_hybrid_medium_streaming_80ms_pc <https://huggingface.co/nvidia/stt_en_fastconformer_hybrid_medium_streaming_80ms_pc>`__ | ||
| - Hybrid | ||
| - ASR, PnC, streaming | ||
| - en | ||
| * - `stt_en_fastconformer_hybrid_medium_streaming_80ms <https://huggingface.co/nvidia/stt_en_fastconformer_hybrid_medium_streaming_80ms>`__ | ||
| - Hybrid | ||
| - ASR, streaming | ||
| - en | ||
| * - `stt_ka_fastconformer_hybrid_transducer_ctc_large_streaming_80ms_pc <https://huggingface.co/nvidia/stt_ka_fastconformer_hybrid_transducer_ctc_large_streaming_80ms_pc>`__ | ||
| - Hybrid | ||
| - ASR, PnC, streaming | ||
| - ka | ||
|
Collaborator
There was a problem hiding this comment. Yeah on Piotr's above point, few know the georgian language code off hand. |
||
| * - `stt_en_fastconformer_hybrid_large_streaming_1040ms <https://catalog.ngc.nvidia.com/orgs/nvidia/teams/nemo/models/stt_en_fastconformer_hybrid_large_streaming_1040ms>`__ | ||
| - Hybrid | ||
| - ASR, streaming | ||
| - en | ||
|
|
||
|
|
||
| FastConformer English Models (Non-Streaming) | ||
| ---------------------------------------------- | ||
|
|
||
| .. list-table:: | ||
| :header-rows: 1 | ||
|
|
||
| * - Model | ||
| - Decoder | ||
| - Capabilities | ||
| - Size | ||
| * - `stt_en_fastconformer_hybrid_large_pc <https://huggingface.co/nvidia/stt_en_fastconformer_hybrid_large_pc>`__ | ||
| - Hybrid | ||
| - ASR, PnC | ||
| - Large | ||
| * - `stt_en_fastconformer_ctc_large <https://huggingface.co/nvidia/stt_en_fastconformer_ctc_large>`__ | ||
| - CTC | ||
| - ASR | ||
| - Large | ||
| * - `stt_en_fastconformer_ctc_xlarge <https://huggingface.co/nvidia/stt_en_fastconformer_ctc_xlarge>`__ | ||
| - CTC | ||
| - ASR | ||
| - XLarge | ||
| * - `stt_en_fastconformer_ctc_xxlarge <https://huggingface.co/nvidia/stt_en_fastconformer_ctc_xxlarge>`__ | ||
| - CTC | ||
| - ASR | ||
| - XXLarge | ||
| * - `stt_en_fastconformer_transducer_large <https://huggingface.co/nvidia/stt_en_fastconformer_transducer_large>`__ | ||
| - RNN-T | ||
| - ASR | ||
| - Large | ||
| * - `stt_en_fastconformer_transducer_xlarge <https://huggingface.co/nvidia/stt_en_fastconformer_transducer_xlarge>`__ | ||
| - RNN-T | ||
| - ASR | ||
| - XLarge | ||
| * - `stt_en_fastconformer_transducer_xxlarge <https://huggingface.co/nvidia/stt_en_fastconformer_transducer_xxlarge>`__ | ||
| - RNN-T | ||
| - ASR | ||
| - XXLarge | ||
| * - `stt_en_fastconformer_tdt_large <https://huggingface.co/nvidia/stt_en_fastconformer_tdt_large>`__ | ||
| - TDT | ||
| - ASR | ||
| - Large | ||
|
|
||
|
|
||
| FastConformer Multilingual Models | ||
| ---------------------------------- | ||
|
|
||
| .. list-table:: | ||
| :header-rows: 1 | ||
|
|
||
| * - Model | ||
|
Collaborator
There was a problem hiding this comment. I'd move all fastconformers underneath parakeet. This'll just lead to confusion.
Collaborator
There was a problem hiding this comment. I think it's OK, the concept here is that
Collaborator
There was a problem hiding this comment. ehhh, i think our branding efforts are causing confusion, especially now Nemotron Speech is a thing. In the technical docs there should be clear understanding that these are the same architectures. The naming aspect can be left up to marketing but for devs it should be clear that fastcomformer and parakeet are largely equivalent. |
||
| - Decoder | ||
| - Capabilities | ||
| - Language | ||
| * - `stt_multilingual_fastconformer_hybrid_large_pc_blend_eu <https://catalog.ngc.nvidia.com/orgs/nvidia/teams/nemo/models/stt_multilingual_fastconformer_hybrid_large_pc_blend_eu>`__ | ||
| - Hybrid | ||
| - ASR, PnC | ||
| - Multilingual EU | ||
| * - `stt_de_fastconformer_hybrid_large_pc <https://huggingface.co/nvidia/stt_de_fastconformer_hybrid_large_pc>`__ | ||
| - Hybrid | ||
| - ASR, PnC | ||
| - de | ||
| * - `stt_es_fastconformer_hybrid_large_pc <https://huggingface.co/nvidia/stt_es_fastconformer_hybrid_large_pc>`__ | ||
| - Hybrid | ||
| - ASR, PnC | ||
| - es | ||
| * - `stt_es_fastconformer_hybrid_large_pc_nc <https://huggingface.co/nvidia/stt_es_fastconformer_hybrid_large_pc_nc>`__ | ||
| - Hybrid | ||
| - ASR, PnC | ||
| - es | ||
| * - `stt_fr_fastconformer_hybrid_large_pc <https://huggingface.co/nvidia/stt_fr_fastconformer_hybrid_large_pc>`__ | ||
| - Hybrid | ||
| - ASR, PnC | ||
| - fr | ||
| * - `stt_it_fastconformer_hybrid_large_pc <https://huggingface.co/nvidia/stt_it_fastconformer_hybrid_large_pc>`__ | ||
| - Hybrid | ||
| - ASR, PnC | ||
| - it | ||
| * - `stt_ru_fastconformer_hybrid_large_pc <https://huggingface.co/nvidia/stt_ru_fastconformer_hybrid_large_pc>`__ | ||
| - Hybrid | ||
| - ASR, PnC | ||
| - ru | ||
| * - `stt_ua_fastconformer_hybrid_large_pc <https://huggingface.co/nvidia/stt_ua_fastconformer_hybrid_large_pc>`__ | ||
| - Hybrid | ||
| - ASR, PnC | ||
| - uk | ||
| * - `stt_pl_fastconformer_hybrid_large_pc <https://huggingface.co/nvidia/stt_pl_fastconformer_hybrid_large_pc>`__ | ||
| - Hybrid | ||
| - ASR, PnC | ||
| - pl | ||
| * - `stt_hr_fastconformer_hybrid_large_pc <https://huggingface.co/nvidia/stt_hr_fastconformer_hybrid_large_pc>`__ | ||
| - Hybrid | ||
| - ASR, PnC | ||
| - hr | ||
| * - `stt_be_fastconformer_hybrid_large_pc <https://huggingface.co/nvidia/stt_be_fastconformer_hybrid_large_pc>`__ | ||
| - Hybrid | ||
| - ASR, PnC | ||
| - be | ||
| * - `stt_nl_fastconformer_hybrid_large_pc <https://huggingface.co/nvidia/stt_nl_fastconformer_hybrid_large_pc>`__ | ||
| - Hybrid | ||
| - ASR, PnC | ||
| - nl | ||
| * - `stt_pt_fastconformer_hybrid_large_pc <https://huggingface.co/nvidia/stt_pt_fastconformer_hybrid_large_pc>`__ | ||
| - Hybrid | ||
| - ASR, PnC | ||
| - pt | ||
| * - `stt_fa_fastconformer_hybrid_large <https://huggingface.co/nvidia/stt_fa_fastconformer_hybrid_large>`__ | ||
| - Hybrid | ||
| - ASR | ||
| - fa | ||
| * - `stt_ka_fastconformer_hybrid_large_pc <https://huggingface.co/nvidia/stt_ka_fastconformer_hybrid_large_pc>`__ | ||
| - Hybrid | ||
| - ASR, PnC | ||
| - ka | ||
| * - `stt_hy_fastconformer_hybrid_large_pc <https://huggingface.co/nvidia/stt_hy_fastconformer_hybrid_large_pc>`__ | ||
| - Hybrid | ||
| - ASR, PnC | ||
| - hy | ||
| * - `stt_ar_fastconformer_hybrid_large_pc_v1.0 <https://huggingface.co/nvidia/stt_ar_fastconformer_hybrid_large_pc_v1.0>`__ | ||
| - Hybrid | ||
| - ASR, PnC | ||
| - ar | ||
| * - `stt_ar_fastconformer_hybrid_large_pcd_v1.0 <https://huggingface.co/nvidia/stt_ar_fastconformer_hybrid_large_pcd_v1.0>`__ | ||
| - Hybrid | ||
| - ASR, PnC, diacritization | ||
| - ar | ||
| * - `stt_uz_fastconformer_hybrid_large_pc <https://huggingface.co/nvidia/stt_uz_fastconformer_hybrid_large_pc>`__ | ||
| - Hybrid | ||
| - ASR, PnC | ||
| - uz | ||
| * - `stt_kk_ru_fastconformer_hybrid_large <https://huggingface.co/nvidia/stt_kk_ru_fastconformer_hybrid_large>`__ | ||
| - Hybrid | ||
| - ASR | ||
| - kk, ru | ||
| * - `parakeet-ctc-0.6b-Vietnamese <https://huggingface.co/nvidia/parakeet-ctc-0.6b-Vietnamese>`__ | ||
|
Collaborator
There was a problem hiding this comment. This should go under Parakeet section above |
||
| - CTC | ||
| - ASR | ||
| - vi | ||
|
|
||
|
|
||
| Loading Models | ||
| -------------- | ||
|
|
||
| All models can be loaded via the ``from_pretrained()`` API: | ||
|
|
||
| .. code-block:: python | ||
|
|
||
| import nemo.collections.asr as nemo_asr | ||
|
|
||
| # From HuggingFace (prefix with nvidia/) | ||
| model = nemo_asr.models.ASRModel.from_pretrained("nvidia/parakeet-tdt-0.6b-v2") | ||
|
|
||
| # From NGC (no prefix) | ||
| model = nemo_asr.models.ASRModel.from_pretrained("stt_en_fastconformer_transducer_large") | ||
|
|
||
| To list all available models programmatically: | ||
|
|
||
| .. code-block:: python | ||
|
|
||
| nemo_asr.models.ASRModel.list_available_models() | ||
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
iirc some of these didn't really prioritize PnC no?