xnevar00 commited on
Commit
289ef62
1 Parent(s): 1ea3fde

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +74 -3
README.md CHANGED
@@ -1,3 +1,74 @@
1
- ---
2
- license: apache-2.0
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ ---
4
+
5
+
6
+ # Whisper Medium ATC full
7
+
8
+ This model is a fine-tuned [openai/whisper-medium](https://huggingface.co/openai/whisper-medium) on Czech and English air traffic communication recordings from Czech airport LKKU.
9
+
10
+ It was created as a product of bachelor's thesis at Faculty of Information Technology Brno University of Technology.
11
+
12
+ # Model description
13
+
14
+
15
+ - **Developed by:** Veronika Nevarilova ([@xnevar00](https://huggingface.co/xnevar00)), Igor Szoke ([@iszoke](https://huggingface.co/iszoke))
16
+ - **Shared by:** [BUT FIT](https://huggingface.co/BUT-FIT)
17
+ - **Model type:** Whisper
18
+ - **Languages:** Czech, English
19
+ - **License:** [Apache 2.0](https://www.apache.org/licenses/LICENSE-2.0)
20
+ - **Finetuned from model:** [openai/whisper-medium](https://huggingface.co/openai/whisper-medium)
21
+
22
+
23
+ # Usage
24
+
25
+ ```
26
+ import torch
27
+ from transformers import pipeline
28
+
29
+ audio = "path/to/audio.xx"
30
+ device = "cuda:0" if torch.cuda.is_available() else "cpu"
31
+
32
+ transcribe = pipeline(task="automatic-speech-recognition", model="whisper_full", chunk_length_s=30, device=device)
33
+ transcribe.model.config.forced_decoder_ids = transcribe.tokenizer.get_decoder_prompt_ids(task="transcribe", language="czech")
34
+ print('Transcription:', transcribe(audio)["text"])
35
+ ```
36
+
37
+ # Dataset
38
+
39
+ Training dataset was made of ~5 hours of air traffic communication recordings. Recordings were Czech and English (80:20) and sporadically Slovak.
40
+
41
+ # Output format
42
+
43
+ The model was learned to transcribe every recording word by word. Transcription format of a recording is as follows:
44
+
45
+ Recording: *Oscar Kilo Alpha Bravo Charlie dráha dva nula střední pro přistání volná vítr nula jedna nula stupňů pět uzlů*
46
+
47
+ Transcription: `Oscar Kilo Alpha Bravo Charlie dráha dva nula střední pro přistání volná vítr nula jedna nula stupňů pět uzlů`
48
+
49
+ **Note:** See also model [BUT-FIT/whisper-ATC-czech-short](https://huggingface.co/BUT-FIT/whisper-ATC-czech-short), which abbreviates callsigns and numbers.
50
+
51
+
52
+ # Results
53
+
54
+ The model reached total WER of 14,7 % on unseen Czech and English LKKU recordings. 19.6 % WER was achieved on a testset containing Czech air traffic recordings from other airports, LKPR and LKTB.
55
+
56
+ WER of callsings in LKKU recordings was evaluated to be 6.2 %, while on LKPR and LKTB dataset the model reached 3.6 %.
57
+
58
+ # Training hyperparameters
59
+
60
+ - **learning_rate:** 3e-5
61
+ - **per_device_train_batch_size:** 2
62
+ - **gradient_accumulation_steps:** 8
63
+ - **warmup_ratio:** 0.12
64
+ - **fp16:** True
65
+ - **gradient_checkpointing:** True
66
+ - **evaluation_strategy:** "epoch"
67
+ - **save_strategy:** "epoch"
68
+ - **load_best_model_at_end:** True
69
+ - **metric_for_best_model:** "wer"
70
+ - **num_train_epochs:** 45
71
+
72
+ # Contact
73
+
74
+ For further information don't hesitate to contact Veronika Nevarilova (**[xnevar00@stud.fit.vutbr.cz](xnevar00@stud.fit.vutbr.cz)**) or Igor Szoke (**[szoke@fit.vutbr.cz](szoke@fit.vutbr.cz)**).