Saved in:
| Main Authors: | , |
|---|---|
| Format: | Recurso digital |
| Language: | |
| Published: |
Zenodo
2025
|
| Online Access: | https://doi.org/10.5281/zenodo.15526037 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Table of Contents:
- <h1>Interspeech submission</h1> <h3>Supplementary materials for our submitted paper to Interspeech 2025</h3> <h3>1. Repo structure:</h3> <p>This repository provides supplementary materials for our submission entitled "<strong>Towards High-Quality LLM-Based Data for French Spontaneous Speech Simplification: an Exo-Refinement Approach</strong>" for Interspeech 2025. The main goal is to provide all the codes and prompt templates to generate reproducible results.</p> <p>To this end, we include:</p> <ul> <li>The designed code along with the different generative prompting templates to implement the iterative generation-evaluation workflow and thus produce synthetic simplifications.</li> <li>The set of collected expert-based simplifications, that allowed to evaluate the quality of the generated outputs under different automatic metrics.</li> </ul> <pre><code>interspeech-2025/ ├── README.md ├── data/ │ ├── experts/ │ │ └── expert_simplifications_all.tsv # Collected reference simplifications │ └── orfeo/ │ └── conversations/ # Set of 240 sentences from orfeo-cefc set ├── generate_data_mistral.py # Script to run Mistral-based LLMs ├── generate_data_openai.py # Script to run OpenAI-based LLMs ├── modules/ │ ├── functions_mistral.py │ ├── functions_openai.py │ └── templates/ │ ├── evaluation.py # Evaluation prompts │ ├── generation.py # Generation prompts │ └── topeval.py # Top-evaluator prompt └── requirements.txt </code></pre> <h3>2. Installation:</h3> <p><strong>Prerequisites</strong>:</p> <ul> <li>Python 3.10.8 or above.</li> <li>Obtain a MistralAI API key: https://console.mistral.ai/api-keys/ (alternatively, run LLMs locally).</li> <li>Obtain an OpenAI API key: https://openai.com/index/openai-api/.</li> </ul> <pre><code>$ git clone git@github.com:lormaechea/interspeech-2025.git $ cd interspeech-2025 $ pip install -e requirements.txt </code></pre> <h3>3. How to use:</h3> <pre><code>$ python generate_data_mistral.py # To generate Mistral-based synthetic simplifications $ python generate_data_openai.py # To generate OpenAI-based synthetic simplifications </code></pre> <h3>4. Authors</h3> <p><!-- If you find this repository helpful, feel free to cite our publication [Towards High-Quality LLM-Based Data for French Spontaneous Speech Simplification: an Exo-Refinement Approach](https://to_appear):--></p> <pre><code>@inproceedings{ormaechea-2025-speech-simplification, title = {Towards High-Quality LLM-Based Data for French Spontaneous Speech Simplification: an Exo-Refinement Approach}, author = {Lucía Ormaechea, Nikos Tsourakis, Pierrette Bouillon, Benjamin Lecouteux and Didier Schwab}, booktitle = {Proc. Interspeech 2025}, year = {2025}, location = {Rotterdam, Netherlands}, url = {To appear}, } </code></pre> <p><strong>Contact person</strong>: <a href="https://luciaormaechea.com/">Lucía Ormaechea</a>, <a href="mailto:lucia.ormaecheagrijalba@unige.ch">lucia.ormaecheagrijalba@unige.ch</a></p> <p>If you have further questions, don't hesitate to send us an email.</p>