ReportAId/medwhisper-large-v3-ita-ct2

Model Card: MedWhisper Large ITA (Ctranslate2 / Faster-Whisper)

This repository contains the CTranslate2 (Faster-Whisper) export of MedWhisper Large ITA, optimized for fast and memory-efficient inference.

Model Details

• —This model is functionally equivalent to MedWhisper Large ITA, since the model weights are identical to the original checkpoints, but exported with ctranslate2 for faster inference.
• —All training details, dataset description, and results are available in the original model card
• —Use this repo if your focus is deployment and near real-time transcription

Model Description

• —Developed by: ReportAId AI Team
• —Model type: Automatic Speech Recognition (ASR)
• —Language(s): Italian
• —License: Private dataset, model released for research and experimentation purposes
• —Finetuned from: openai/whisper-large-v3-turbo

How we exported this model for Faster-Whisper

This repository was created by converting the base model ReportAId/medwhisper-large-v3-ita into the ctranslate2 format, which is directly supported by Faster-Whisper.

To export the model:

• —Install ctranslate2 and Faster-Whisper

With pip:

pip install faster-whisper ctranslate2

With uv:

uv add faster-whisper ctranslate2

With Poetry:

poetry add faster-whisper ctranslate2

• —Run:

from ctranslate2.converters import TransformersConverter

path_sample='path/to/model'

model=TransformersConverter("ReportAId/medwhisper-large-v3-ita",copy_files=['preprocessor_config.json','tokenizer.json',])

model.convert(path_sample)

Uses

Direct Use

• —Automatic transcription in Italian
• —Use in reporting systems, meeting transcription, voice-to-text

Inference speed

On 4 CPU cores, transcription runs in ~50–70% of the input audio duration (RTF ≈ 0.5–0.7), making it suitable for near real-time transcription. For example, a 1-minute audio file is processed in ~30–42 seconds.

How to Get Started with the Model

from faster_whisper import WhisperModel

model=WhisperModel('ReportAId/medwhisper-large-v3-ita-ct2')


segments, info = model.transcribe('audio.mp3')

for segment in segments:
    print("[%.2fs -> %.2fs] %s" % (segment.start, segment.end, segment.text))

Model Card Contact

• —Team: ReportAId