Hugging Face's logo

aether-raid
/
astra-atc-models

Automatic Speech Recognition
NeMo
Safetensors
English
parakeet
whisper
qwen3
ctranslate2
text-generation
air-traffic-control
atc
singapore
military
Model card Files
xet
 Community
RanenSim commited on
Commit
9dd7a6b
·
1 Parent(s): 01f9953

feat: update model

Browse files
Files changed (3) hide show
  1. ASR/parakeet/README.md +4 -3
  2. ASR/parakeet/model.ckpt +1 -1
  3. README.md +2 -2
ASR/parakeet/README.md CHANGED
@@ -23,7 +23,7 @@ model-index:
23
  metrics:
24
  - name: Validation WER
25
  type: wer
26
- value: 1.14
27
  ---
28
 
29
  # Parakeet-TDT 0.6B v2 - Singapore Military ATC
@@ -34,8 +34,8 @@ Fine-tuned NVIDIA Parakeet-TDT 0.6B v2 for Singapore Air Force air traffic contr
34
 
35
  | Checkpoint | Validation WER | Notes |
36
  |------------|----------------|-------|
37
- | `model.ckpt` | **1.14%** | Best checkpoint |
38
- | `epoch=54-val_wer=0.0245-last.ckpt` | 2.45% | Final checkpoint, not published here |
39
 
40
  ## Model Details
41
 
@@ -47,6 +47,7 @@ Fine-tuned NVIDIA Parakeet-TDT 0.6B v2 for Singapore Air Force air traffic contr
47
  | Format | Raw `.ckpt` checkpoint + tokenizer artifacts |
48
  | Checkpoint size | 7.0 GB |
49
  | Domain | Singapore military ATC (Tengah WSAT, Paya Lebar WSAP) |
 
50
 
51
  ## Included Files
52
 
 
23
  metrics:
24
  - name: Validation WER
25
  type: wer
26
+ value: 0.72
27
  ---
28
 
29
  # Parakeet-TDT 0.6B v2 - Singapore Military ATC
 
34
 
35
  | Checkpoint | Validation WER | Notes |
36
  |------------|----------------|-------|
37
+ | `model.ckpt` | **0.72%** | Best checkpoint from epoch 76 |
38
+ | `epoch=100-val_wer=0.0073-last.ckpt` | 0.73% | Final checkpoint, not published here |
39
 
40
  ## Model Details
41
 
 
47
  | Format | Raw `.ckpt` checkpoint + tokenizer artifacts |
48
  | Checkpoint size | 7.0 GB |
49
  | Domain | Singapore military ATC (Tengah WSAT, Paya Lebar WSAP) |
50
+ | Training data | Clean originals plus mild ATC radio, speed, and stress augmentations |
51
 
52
  ## Included Files
53
 
ASR/parakeet/model.ckpt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:8f2ff18de72abdbe7fbde1dc50be1cf5dd156f58d258940c25d33db14cc23bfc
3
  size 7415423311
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:abcd3d1c223ba5754a4b00e5d7e7406b9f9cbe2d731b519417f5bbe8b34febf8
3
  size 7415423311
README.md CHANGED
@@ -50,7 +50,7 @@ Fine-tuned NeMo Parakeet model for Singapore military ATC speech. Published as a
50
 
51
  | Metric | Value |
52
  |--------|-------|
53
- | Validation WER | **1.14%** |
54
  | Base model | `nvidia/parakeet-tdt-0.6b-v2` |
55
  | Size | 7.0 GB |
56
  | Runtime | `nemo_toolkit[asr]` |
@@ -94,7 +94,7 @@ Singapore military ATC covering Tengah and Paya Lebar operations, military phras
94
  | ct2_run6 | 0.40% | jacktol/whisper-large-v3-finetuned-for-ATC | +augmentation, weight decay |
95
  | ct2_run7 | 0.24% | jacktol/whisper-large-v3-finetuned-for-ATC | Frozen encoder, +50 real recordings |
96
  | ct2_run8 | 0.66% | openai/whisper-large-v3 | Full retrain from base, enhanced augmentation |
97
- | parakeet_atc | 1.14% | nvidia/parakeet-tdt-0.6b-v2 | NeMo fine-tune, best checkpoint at epoch 20 |
98
 
99
  ### LLM
100
 
 
50
 
51
  | Metric | Value |
52
  |--------|-------|
53
+ | Validation WER | **0.72%** |
54
  | Base model | `nvidia/parakeet-tdt-0.6b-v2` |
55
  | Size | 7.0 GB |
56
  | Runtime | `nemo_toolkit[asr]` |
 
94
  | ct2_run6 | 0.40% | jacktol/whisper-large-v3-finetuned-for-ATC | +augmentation, weight decay |
95
  | ct2_run7 | 0.24% | jacktol/whisper-large-v3-finetuned-for-ATC | Frozen encoder, +50 real recordings |
96
  | ct2_run8 | 0.66% | openai/whisper-large-v3 | Full retrain from base, enhanced augmentation |
97
+ | parakeet_atc | 0.72% | nvidia/parakeet-tdt-0.6b-v2 | NeMo fine-tune with ATC radio augmentation, best checkpoint at epoch 76 |
98
 
99
  ### LLM
100