---
license: llama3.2
language:
- en
base_model: twangodev/rasr-parakeet-v1
base_model_relation: finetune
tags:
- automatic-speech-recognition
- asr
- atc
- air-traffic-control
- aviation
- parakeet
- nemo
- fastconformer
- tdt
- finetuned
- built-with-llama
datasets:
- twangodev/tartanaviation-atc-adsb-utterances
- twangodev/tartanaviation-atc-labels
- twangodev/tartanaviation-atc-adsb
- jlvdoorn/atco2-asr
- jlvdoorn/atco2-asr-atcosim
metrics:
- wer
- cer
library_name: nemo
pipeline_tag: automatic-speech-recognition
model-index:
- name: rasr-parakeet-v2
  results:
  - task:
      type: automatic-speech-recognition
      name: Speech-to-Text
    dataset:
      name: liveATC (SAadettin-BERber/liveATC_merged)
      type: SAadettin-BERber/liveATC_merged
      split: train
    metrics:
    - type: wer
      value: 0.2513
      name: Word Error Rate
    - type: cer
      value: 0.1744
      name: Character Error Rate
  - task:
      type: automatic-speech-recognition
      name: Speech-to-Text
    dataset:
      name: ATCO2 (jlvdoorn/atco2-asr validation)
      type: jlvdoorn/atco2-asr
      split: validation
    metrics:
    - type: wer
      value: 0.1277
      name: Word Error Rate
    - type: cer
      value: 0.0840
      name: Character Error Rate
---

# rasr-parakeet-v2

ATC ASR model targeting US air traffic control. A continued finetune of [`twangodev/rasr-parakeet-v1`](https://huggingface.co/twangodev/rasr-parakeet-v1) on real US ATC audio from the [TartanAviation](https://theairlab.org/tartanaviation/) corpus (CMU AirLab). v2 of the [rasr](https://github.com/twangodev/rasr) toolkit.

v1 was strong on European ATC but weak on US ATC, since its only real-audio anchor was European. v2 adds about 29 hours of real US tower audio to close that gap.

## Headline

| Metric | v2 | v1 |
|---|---|---|
| **liveATC WER (US)** | **0.251** | 0.311 |
| **liveATC CER (US)** | **0.174** | 0.207 |
| ATCO2 val WER (EU) | 0.128 | 0.125 |
| ATCO2 val CER (EU) | 0.084 | 0.078 |

About 19% relative WER reduction on US ATC (liveATC) over v1, with European performance held flat. Both rows use `rasr eval run` on the same references, so the comparison is direct. One caveat: readback numeric WER regressed (see Limitations).

## Quick start

```python
import nemo.collections.asr as nemo_asr

model = nemo_asr.models.ASRModel.from_pretrained("twangodev/rasr-parakeet-v2")
result = model.transcribe(["atc_clip.wav"])
print(result[0].text)
```

Or via the rasr eval toolkit:

```bash
pip install rasr
rasr eval run \
  -m nemo:hf://twangodev/rasr-parakeet-v2 \
  -d hf:SAadettin-BERber/liveATC_merged:train \
  --language en --batch-size 16
```

## Architecture

- **Base**: [`nvidia/parakeet-tdt-0.6b-v3`](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3) (FastConformer encoder, TDT decoder, 0.6B params)
- **Parent**: [`twangodev/rasr-parakeet-v1`](https://huggingface.co/twangodev/rasr-parakeet-v1) (this model is a continued finetune of v1)
- **Tokenizer**: kept from base, SentencePiece BPE 8192 tokens
- **Sample rate**: 16 kHz mono
- **Max input duration**: 18 seconds (longer inputs may degrade; TDT joint memory)

## Training data

v2 starts from v1's weights and continues training on the mix below. v1's own training (synthetic transcripts plus a European real anchor) is inherited in the weights; see the [v1 card](https://huggingface.co/twangodev/rasr-parakeet-v1).

| Source | Type | Role |
|---|---|---|
| [TartanAviation](https://huggingface.co/datasets/twangodev/tartanaviation-atc-adsb-utterances), high inter-model agreement subset | Real US ATC (KAGC tower) | US training audio: 35,240 clips, 28.8 h |
| [`jlvdoorn/atco2-asr-atcosim`](https://huggingface.co/datasets/jlvdoorn/atco2-asr-atcosim) (train) | Real EU ATC plus simulator | European rehearsal |
| [`jlvdoorn/atco2-asr`](https://huggingface.co/datasets/jlvdoorn/atco2-asr) (train) | Real EU ATC | European rehearsal |

TartanAviation ships no transcripts. The training targets are pseudo labels from a weighted ROVER fusion of three ASR models (`nvidia/parakeet-tdt-0.6b-v2`, `nvidia/canary-qwen-2.5b`, `jlvdoorn/whisper-large-v3-atco2-asr`) plus ADS-B callsign snapping, filtered to the highest inter-model agreement ([`twangodev/tartanaviation-atc-labels`](https://huggingface.co/datasets/twangodev/tartanaviation-atc-labels)). On a human-reviewed sample this subset is about 98% acceptable, but these are not human labels. Evaluation clips are held out by date.

TartanAviation components used:

- [`twangodev/tartanaviation-atc-adsb-utterances`](https://huggingface.co/datasets/twangodev/tartanaviation-atc-adsb-utterances): segmented US ATC audio (16 kHz, with per-clip ADS-B tracks).
- [`twangodev/tartanaviation-atc-labels`](https://huggingface.co/datasets/twangodev/tartanaviation-atc-labels): the pseudo-label transcripts and agreement scores used as supervised targets.
- [`twangodev/tartanaviation-atc-adsb`](https://huggingface.co/datasets/twangodev/tartanaviation-atc-adsb): ADS-B trajectory data, used for the callsign snapping in the labels.

## Llama 3.2 attribution

This model is **"Built with Llama"** under the [Llama 3.2 Community License](https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/LICENSE). v2 is a continued finetune of v1, whose supervised targets were ATC dialogue transcripts generated by Llama 3.2; v2 is therefore a derivative of those Llama-trained weights. The new US targets in v2 were not generated by Llama (they are the pseudo labels described above).

## TartanAviation attribution

US training audio is from the **TartanAviation** dataset (CMU AirLab, **CC-BY-4.0**). Please credit CMU AirLab and cite:

```bibtex
@article{patrikar2024tartanaviation,
  title={TartanAviation: Image, Speech, and ADS-B Trajectory Datasets for Terminal Airspace Operations},
  author={Jay Patrikar and Joao Dantas and Brady Moon and Milad Hamidi and Sourish Ghosh and Nikhil Keetha and Ian Higgins and Atharva Chandak and Takashi Yoneyama and Sebastian Scherer},
  year={2024},
  eprint={2403.03372},
  archivePrefix={arXiv},
  primaryClass={cs.LG},
  url={https://arxiv.org/pdf/2403.03372.pdf}
}
```

## Training recipe

Continued finetune from v1 (`rasr-parakeet-v1.nemo`). Recipe: `training_recipe.yaml` (this repo).

| Hyperparameter | Value |
|---|---|
| Initialize from | `twangodev/rasr-parakeet-v1` (continued finetune, not from base) |
| Optimizer | AdamW, betas (0.9, 0.98), weight_decay 1e-3 |
| Learning rate | 5e-5 |
| Schedule | CosineAnnealing, warmup 300 steps, min_lr 1e-6 |
| Batch size | 48 |
| Precision | bf16-mixed |
| Max steps | 10,000 |
| Augmentation | SpecAugment (default), speed perturb 0.95 to 1.05 |
| Max audio duration | 18.0 s |
| Mixing | weighted manifest concat (TartanAviation x1, ATCO2 train x1, ATCO2+ATCOSIM train x1) |
| Hardware | NVIDIA RTX PRO 6000 Blackwell (96 GB) |
| Wall clock | about 1 hour |

## Strengths

- **Lower US ATC WER**: about 19% relative WER and 16% relative CER reduction on liveATC vs v1.
- **Real US adaptation**: held-out US in-domain WER drops from 0.13 to 0.05; the model learns real US tower phraseology instead of substituting European or airline priors.
- **European performance maintained**: ATCO2 val WER 0.125 to 0.128, no catastrophic forgetting.
- **Small footprint**: 0.6B params, fits in 4 GB VRAM, same speed and size as v1.

## Limitations

1. **Numeric readback regression.** Digit WER got worse than v1: US numeric 0.371 to 0.421, EU numeric 0.050 to 0.064 (`rasr eval run`). v1's strength was numeric accuracy, and v2 trades some of it for the overall gain. A second normalization pipeline showed numeric roughly flat rather than worse, so the magnitude is uncertain, but it did not improve. For digit-critical use, evaluate v1 as well and layer hot-word or grammar constraints on number fields.
2. **Narrow US base.** The US audio is 28.8 h from a single general-aviation tower (KAGC). It generalized to liveATC well, but expect weaker results on US airspace very different from a GA tower (busy approach or center).
3. **US eval is relative.** liveATC labels are auto-generated, so the 19% is a relative v1 to v2 delta on identical references, not an absolute benchmarked WER. No fully public human-labeled US ATC ASR benchmark currently exists.
4. **Pseudo-label training.** US supervised targets are model-generated, not human labels, so residual systematic labeler errors (especially digits) may be inherited.
5. **Inherited v1 biases.** Operator substitution and European-prior artifacts from v1 may persist on out-of-distribution audio.

## Recommended usage

- **US ATC**: v2 is the better default for overall transcription. Use inference-time hot-word biasing against a known callsign and airport-name vocabulary (NeMo TDT supports it via `change_decoding_strategy()`).
- **Numeric or readback-critical content**: compare against v1, and consider grammar or format constraints on digit fields.
- **European ATC**: v1 and v2 are about equivalent; either works.
- **Safety-critical applications**: this is a research and development checkpoint, not a safety-certified ATC transcription system. Always layer confidence-based rejection.

## Citation

```bibtex
@software{rasr,
  author = {Ding, James},
  title = {rasr: ATC ASR finetuning toolkit},
  url = {https://github.com/twangodev/rasr},
  year = {2026}
}
```

Base model:

```bibtex
@misc{parakeet-tdt,
  author = {NVIDIA},
  title = {Parakeet-TDT-0.6B-v3},
  url = {https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3}
}
```

US training audio (TartanAviation) and Llama 3.2 (inherited from v1): see the attribution sections above.

## License

Released under the **[Llama 3.2 Community License](https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/LICENSE)** ("Built with Llama"). This is the binding upstream license: v2 is a continued finetune of v1, whose training transcripts were generated by Llama 3.2, so the weights are treated as a derivative work of Llama Materials.

This model also inherits attribution and use requirements from its other parents:

- **Parakeet-TDT-0.6B-v3** ([CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/), NVIDIA), base architecture
- **TartanAviation** ([CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/), CMU AirLab), real US training audio; credit CMU AirLab and cite the paper above
- **ATCO2 corpus** (CC-BY-4.0), European real anchor (train split)
- **ATCOSIM corpus** (research use; see [source](https://www.spsc.tugraz.at/databases-and-tools/atcosim-air-traffic-control-simulation-speech-corpus.html))
- **radiotalk-us-audio-tada-noisy** (Llama 3.2 Community License), inherited from v1

To redistribute or deploy:
1. Include a copy of the Llama 3.2 Community License.
2. Display "Built with Llama" in your product, user interface, or about page.
3. Provide CC-BY-4.0 attribution to CMU AirLab (TartanAviation), NVIDIA (Parakeet), and the ATCO2 authors.
4. Comply with the [Llama 3.2 Acceptable Use Policy](https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/USE_POLICY.md).
5. If your service exceeds 700M monthly active users, request a separate commercial license from Meta.

This is not legal advice. If you are deploying this model commercially or at scale, consult a lawyer regarding the interaction of the upstream licenses.