Taraneh-WMedium: Persian Automatic Speech Recognition

Taraneh-WMedium is the Whisper-Medium member of Taraneh, a family of Persian ASR models fine-tuned on about 1,560 hours of speech. The base checkpoint is openai/whisper-medium.

License

MIT, the same license as OpenAI Whisper. Commercial use is allowed, including in products and hosted services. Keep this notice and the original Whisper MIT notice with the weights.

Paper

Aref Farhadipour. Taraneh-ASR: Scaling Persian Speech Recognition and Forced Alignment Across Singing and Conversational Domains. University of Zurich.

• PDF in the code repository: https://github.com/areffarhadi/persian-asr/blob/main/ArefFarhadi-TaranehASR.pdf
• Code, manifests, normalization, and inference: https://github.com/areffarhadi/persian-asr

Load this checkpoint with Transformers (WhisperForConditionalGeneration). CTranslate2 is the fast path in the GitHub scripts.

Forced alignment

This checkpoint can return word times as well as a transcript. In the GitHub scripts, --task asr writes the transcript and --task asr+fa also writes alignments.jsonl.

The readout uses decoder cross-attention heads L19/H8 and L16/H12, a median filter of width 3, column normalization, dynamic time warping, and a −35 dB trim of the first-word start and last-word end. On the FarsDat gold set this model reaches 37.8 ms mean absolute error. Heads and the command are in the GitHub README.

Results

Word error rate / character error rate (%) after the unified Persian text normalization. PSRB and VisualEars were unseen in training.

FLEURS	Common Voice	YouTube	Music	Media	PSRB	VisualEars
4.16 / 1.52	3.47 / 1.67	14.29 / 8.26	22.79 / 10.92	14.97 / 8.99	19.14 / 8.73	6.23 / 1.89

The full comparison of all five Taraneh models is in the paper and in the GitHub README.