Code For Speech Generation/README.md

Speech generation code (DEAF)

Scripts used to build the BSC and SIC speech datasets for DEAF: A Benchmark for Diagnostic Evaluation of Acoustic Faithfulness in Audio Language Models.

BSC pipeline

1. BSC/edgeTTS.py — Synthesize speech from text with Edge TTS.
2. BSC/mp3_to_wav.py — Convert synthesized speech and background audio (e.g. MP3) to WAV at 16 kHz.
3. BSC/addNoise.py — Mix speech with background noise to produce the final BSC samples.

Run these steps in order from the repository root or adjust paths to match your local layout.

SIC pipeline

• SIC/SIC_audio_generation.py — Generate speech from text for the SIC subset (single script end-to-end for this track). Speech synthesis uses the ElevenLabs Python SDK.

Reference (tooling): ElevenLabs. 2024. ElevenLabs Python SDK. https://github.com/elevenlabs/elevenlabs-python

Requirements

Install dependencies used by the scripts you run (for example edge-tts, pydub, soundfile, numpy, scipy, elevenlabs for the SIC pipeline, and any others imported in each file). Pin versions in your own requirements.txt if you publish this folder as a standalone project.