---
license: cc-by-2.0
language:
- it
---

# Model Card for Model ID

<!-- Provide a quick summary of what the model is/does. -->

This modelcard aims to be a base template for new models. It has been generated using [this raw template](https://github.com/huggingface/huggingface_hub/blob/main/src/huggingface_hub/templates/modelcard_template.md?plain=1).

## Model Details

The model has 31,536,128 trainable parameters

### Model Description

<!-- Provide a longer summary of what this model is. -->

Model trained using Early Exit architecture: 12 conformer layers, 6 CTC decoders. 
The model has been generated by averaging from epoch 16 to epoch 26. 
This model can handle only speech signals sampled at 16 kHz.

## Uses

<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->

To be used for ASR: code for using the model available at https://github.com/SpeechTechLab/early-exit-transformer

### Recommendations

<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.

## How to Get Started with the Model

Use the code at https://github.com/SpeechTechLab/early-exit-transformer.


## Training Details

decoder_mode='ctc', model_type='early_conformer', bpe=True

distill=False, language_model=None, language_model_dict=None, avg_model_start=0, avg_model_end=5

max_len=2000, d_model=256, n_enc_layers_per_exit=2, n_enc_exits=6, n_dec_layers=6, n_heads=8

d_feed_forward=2048, depthwise_kernel_size=31, max_utterance_length=600, sample_rate=16000

n_fft=512, win_length=320, hop_length=160, n_mels=80

src_pad_idx=0, trg_pad_idx=126, trg_sos_idx=1, trg_eos_idx=2, enc_voc_size=256, dec_voc_size=256

sp=<sentencepiece.SentencePieceProcessor=;'cv.bpe-256.model' lexicon='cv-bpe-256.lex', tokens='cv-bpe-256.tok')


### Training Data

<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->

Common Voice (Italian) [~410h],

MultiLingual LibriSpeech (Italian) [~247h],

VoxPopuli (Italian) [~87h],

You Tube Commons (Italian) [~1580h]


### Training Procedure

<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->

47 epochs on CV followed by  80 epochs on CV+MLS+Voxpopuli followed by 5 epochs on YPT+CV+MLS+Voxpopuli


#### Training Hyperparameters

shuffle=True, batch_size=64, n_batch_split=8, drop_prob=0.1, init_lr=1e-05, adam_eps=1e-09, weight_decay=0.0001, warmup=[trining dataset size], clip=1.0

#### Speeds, Sizes, Times [optional]

<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->

[More Information Needed]

## Evaluation (%WER)

<!-- This section describes the evaluation protocols and provides the results. -->


   | MLS         | Voxpopuli | CV      | 
   |-----------  |---------  | ------- |
   |  17.66      |   19.69   | 19.42     


### Testing Data, Factors & Metrics

#### Testing Data

<!-- This should link to a Dataset Card if possible. -->

[More Information Needed]

#### Factors

<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->

[More Information Needed]

#### Metrics

<!-- These are the evaluation metrics being used, ideally with a description of why. -->

[More Information Needed]

### Results

[More Information Needed]

#### Summary



## Model Examination [optional]

<!-- Relevant interpretability work for the model goes here -->

[More Information Needed]

## Environmental Impact

<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->

Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).

- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]

## Technical Specifications [optional]

### Model Architecture and Objective

[More Information Needed]

### Compute Infrastructure

FBK-digis cluster

#### Hardware

device=device(type='cuda', index=0, CUDA Version: 12.5) GPU quadro RTX50000

#### Software

[More Information Needed]

## Citation [optional]

<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->

G. A. Wright, U. Cappellazzo, S. Zaiem, D. Raj, L. O. Yang, D. Falavigna, M. N. Ali, and A. Brutti, “Training early-exit architectures for automatic speech recognition: 
Fine-tuning pre-trained models or training from scratch,” in Proc. of ICASSP Workshops, 2024, pp. 685–689 (https://arxiv.org/abs/2309.09546)

Maxence Lasbordes, Daniele Falavigna, Alessio Brutti, “Splitformer: An improved early-exit architecture for automatic speech recognition on edge devices”, 
Proc. of EUSIPCO 2025 (https://arxiv.org/abs/2506.18035)

Mohamed Nabih Ali, Alessio Brutti, Daniele Falavigna, Federating Dynamic Models using Early-Exit Architectures for Automatic Speech Recognition on Heterogeneous Clients. 
To appear on "Progress in Artificial Intelligence" (https://arxiv.org/abs/2405.17376)


**BibTeX:**

[More Information Needed]

**APA:**

[More Information Needed]

## Glossary [optional]

<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->

[More Information Needed]

## More Information [optional]

[More Information Needed]

## Model Card Authors [optional]

[More Information Needed]

## Model Card Contact

[More Information Needed]