RASMUS/Finnish-ASR-Canary-v2
02.2k
1Speech Data Explorer2--------------------3 4[Dash](https://plotly.com/dash/)-based tool for interactive exploration of ASR/TTS datasets.5 6Features:7- dataset's statistics (alphabet, vocabulary, duration-based histograms)8- navigation across dataset (sorting, filtering)9- inspection of individual utterances (waveform, spectrogram, audio player)10- errors' analysis (Word Error Rate, Character Error Rate, Word Match Rate, Mean Word Accuracy, diff)11- comparison of two ASR models using interactive word-level accuracy plot 12 13Please make sure that requirements are installed. Then run:14```15python data_explorer.py path_to_manifest.json16```17 18To compare word-level accuracy of two ASR models:19```20python data_explorer.py path_to_manifest.json -nc pred_text_{model_1_name} pred_text_{model_2_name}21```22 23JSON manifest file should contain the following fields:24- "audio_filepath" (path to audio file)25- "duration" (duration of the audio file in seconds)26- "text" (reference transcript)27 28Errors' analysis requires "pred_text" (ASR transcript) for all utterances.29"Visual comparison requires two or more "pred_text_{model_name}" fields."30 31Any additional field will be parsed and displayed in 'Samples' tab.32 33 3435