Fasih-TTS-Benchmark: Evaluation Scores for Arabic Text-to-Speech Model

Evaluation audio and objective scores for the Fasih-TTS-V1 Arabic text-to-speech model. The dataset contains model outputs paired with reference text, Whisper transcriptions, and per-clip Word Error Rate (WER) and Character Error Rate (CER) scores. It was created by NightPrince and last updated on July 7, 2026.

Use Cases

• Benchmarking Arabic TTS model performance based on Word Error Rate (WER) and Character Error Rate (CER) scores
• Evaluating transcription accuracy of speech models based on Whisper-large-v3 transcriptions
• Analyzing model consistency and output quality across different splits like 'silma_msa' and 'consistency'
• Comparing TTS model outputs against reference text for error analysis

Strengths

• Includes per-clip WER and CER scores, providing granular evaluation metrics
• Contains multiple evaluation splits with reported mean CER, such as 2.0% for the 'silma_msa' split
• Pairs model audio outputs with reference text and Whisper transcriptions for multi-faceted analysis

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download
• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment

Provenance

Source

huggingface

Collection Method

Likely generated as part of evaluating the Fasih-TTS-V1 model's outputs.

Freshness

Last updated 2026-07-07 20:26:42