Hugging Face's logo

Edmon02
/
speecht5_finetuned_hy

Text-to-Speech
Transformers
Safetensors
Armenian
English
Dutch
speecht5
text-to-audio
archived
Model card Files
xet
 Community
Edmon02 commited on
Commit
bf08481
·
verified ·
1 Parent(s): 9f99be3

docs: advanced model card (Armenian SpeechT5)

Browse files
Files changed (1) hide show
  1. README.md +37 -51
README.md CHANGED
@@ -1,71 +1,57 @@
1
  ---
2
- inference: false
 
 
 
3
  license: mit
4
  base_model: microsoft/speecht5_tts
5
- tags:
6
- - generated_from_trainer
7
- model-index:
8
- - name: speecht5_finetuned_hy
9
- results: []
10
- language:
11
- - hy
12
- - en
13
- - nl
14
  datasets:
15
- - mozilla-foundation/common_voice_11_0
 
 
 
 
16
  pipeline_tag: text-to-speech
 
17
  ---
18
 
19
- <!-- This model card has been generated automatically according to the information the Trainer had access to. You
20
- should probably proofread and complete it, then remove this comment. -->
21
-
22
- # speecht5_finetuned_hy
23
-
24
- This model is a fine-tuned version of [microsoft/speecht5_tts](https://huggingface.co/microsoft/speecht5_tts) on the None dataset.
25
- It achieves the following results on the evaluation set:
26
- - Loss: 0.4785
27
-
28
- ## Model description
29
 
30
- More information needed
31
 
32
- ## Intended uses & limitations
33
 
34
- More information needed
35
 
36
- ## Training and evaluation data
37
 
38
- More information needed
 
 
 
 
 
39
 
40
- ## Training procedure
41
 
42
- ### Training hyperparameters
 
 
43
 
44
- The following hyperparameters were used during training:
45
- - learning_rate: 1e-05
46
- - train_batch_size: 4
47
- - eval_batch_size: 2
48
- - seed: 42
49
- - gradient_accumulation_steps: 8
50
- - total_train_batch_size: 32
51
- - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
52
- - lr_scheduler_type: linear
53
- - lr_scheduler_warmup_steps: 125
54
- - training_steps: 1000
55
 
56
- ### Training results
 
 
 
57
 
58
- | Training Loss | Epoch | Step | Validation Loss |
59
- |:-------------:|:-----:|:----:|:---------------:|
60
- | 0.5547 | 2.04 | 250 | 0.4983 |
61
- | 0.525 | 4.07 | 500 | 0.4864 |
62
- | 0.52 | 6.11 | 750 | 0.4812 |
63
- | 0.5286 | 8.15 | 1000 | 0.4785 |
64
 
 
 
 
 
65
 
66
- ### Framework versions
67
 
68
- - Transformers 4.35.2
69
- - Pytorch 2.1.0+cu121
70
- - Datasets 2.16.0
71
- - Tokenizers 0.15.0
 
1
  ---
2
+ language:
3
+ - hy
4
+ - en
5
+ - nl
6
  license: mit
7
  base_model: microsoft/speecht5_tts
 
 
 
 
 
 
 
 
 
8
  datasets:
9
+ - mozilla-foundation/common_voice_11_0
10
+ tags:
11
+ - archived
12
+ - speecht5
13
+ - text-to-speech
14
  pipeline_tag: text-to-speech
15
+ library_name: transformers
16
  ---
17
 
18
+ # Archived baseline: `speecht5_finetuned_hy`
 
 
 
 
 
 
 
 
 
19
 
20
+ First fine-tune of [microsoft/speecht5_tts](https://huggingface.co/microsoft/speecht5_tts) on **Common Voice 11** (multilingual tags: hy, en, nl). Vocab **81**.
21
 
22
+ ## Status
23
 
24
+ **Archived** — kept for lineage reproducibility. Armenian production work moved to HyVoxPopuli-based checkpoints.
25
 
26
+ ## Successor chain
27
 
28
+ ```
29
+ speecht5_finetuned_hy (this repo)
30
+ → speecht5_finetuned_voxpopuli_nl
31
+ → speecht5_finetuned_voxpopuli_hy ← production
32
+ → TTS_NB_2 ← active training
33
+ ```
34
 
35
+ ## Evaluation (historical)
36
 
37
+ | Step | Validation loss |
38
+ |------|-----------------|
39
+ | 1000 | 0.4785 |
40
 
41
+ ## Training hyperparameters
 
 
 
 
 
 
 
 
 
 
42
 
43
+ - Learning rate: 1e-5
44
+ - Effective batch size: 32
45
+ - Training steps: 1000
46
+ - Transformers 4.35.2, PyTorch 2.1.0
47
 
48
+ ## Use instead
 
 
 
 
 
49
 
50
+ | Task | Model |
51
+ |------|--------|
52
+ | Armenian TTS | [Edmon02/speecht5_finetuned_voxpopuli_hy](https://huggingface.co/Edmon02/speecht5_finetuned_voxpopuli_hy) |
53
+ | Fine-tuning | [Edmon02/TTS_NB_2](https://huggingface.co/Edmon02/TTS_NB_2) |
54
 
55
+ ## License
56
 
57
+ MIT