File size: 1,159 Bytes
b18868f
 
 
 
 
 
 
 
 
 
 
54ac444
b18868f
b2abd39
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
---
language:
- en
---

# TalkBank Batchalign CHATWhisper
CHATWhisper is a series of ASR models specifically designed for the task for Language Sample Analysis (LSA) released by the TalkBank project, which delivers superior performance in the analysis of conversational speech transcripts, especially with regards to the analysis of filled pauses, retraicings, and stuttering.

The models are based on `openai/whisper-large-v2` trained using an alpha=32, rank=16 LoRA. We will update the model card with evaluation performance shortly.

## Usage
The models can be used directly as a Whisper-class ASR model following the [same instructions](https://huggingface.co/openai/whisper-base) released by OpenAI. Alternatively, to get the full analysis possible with the model, it is best combined with the TalkBank Batchalign suite of analysis software, [available here](https://github.com/talkbank/batchalign2), using `transcribe` mode with the `--whisper` flag.

## Data
The models are trained with a combination of English Control Protocol samples from [the AphasiaBank corpus](https://aphasia.talkbank.org/) of conversational speech from three seperate corpora.