Text-To-Speech (TTS) Evaluation using objective metrics.
Visualize model performance on function calling tasks
Explore and visualize diverse models
Explore and benchmark visual document retrieval models
Quantize a model for faster inference
Convert Stable Diffusion checkpoint to Diffusers and open a PR
View and submit language model evaluations
Calculate memory needed to train AI models
Upload a machine learning model to Hugging Face Hub
Evaluate open LLMs in the languages of LATAM and Spain.
Search for model performance across languages and benchmarks
Explain GPU usage for model training
Load AI models and prepare your space
The TTSDS Benchmark and Leaderboard is a comprehensive tool designed to evaluate and compare Text-To-Speech (TTS) models using objective metrics. It provides a platform to assess the quality of TTS systems by measuring how closely synthesized speech matches human speech. The leaderboard serves as a central hub to track the performance of various models, enabling easy comparison and fostering advancements in TTS technology.
What metrics does TTSDS Benchmark use?
TTSDS Benchmark primarily uses Mel-Cepstral Distortion (MCD) and Short-Time Objective Intelligibility (STOI) to evaluate TTS models. These metrics are widely accepted for assessing speech synthesis quality.
How do I add my custom TTS model to the leaderboard?
To add your custom model, follow these steps:
How often is the leaderboard updated?
The leaderboard is dynamically updated whenever new models are benchmarked and submitted. However, users are responsible for running the benchmark script and submitting their results to reflect the latest performance of their models.