> For the complete documentation index, see [llms.txt](https://utca.knowledgator.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://utca.knowledgator.com/predictors/openaiwhisperpredictor.md).

# OpenAIWhisperPredictor

Whisper predictor

This predictor is specifically build for OpenAI Whisper model. See more about Whisper:

{% embed url="<https://github.com/openai/whisper>" %}

Subclass of [**Predictor**](/predictors/predictor.md)**.**

## Module: [implementation](/framework-structure.md#implementation).predictors

## Methods and properties

Main methods and properties

***

### <mark style="color:blue;">\_\_init\_\_</mark>

#### Arguments:

* <mark style="color:orange;">**model\_cfg**</mark>**&#x20;(**[**WhisperModelConfig**](#whispermodelconfig)**):** Whisper model configuration. If equals to None, default [**WhisperModelConfig**](#whispermodelconfig) configuration will be used. Defaults to None.
* <mark style="color:orange;">**transcription\_cfg**</mark>**&#x20;(Optional\[**[**WhisperTranscriptionConfig**](#whispertranscriptionconfig)**]):** Transcription configuration. If equals to None, default [**WhisperTranscriptionConfig**](#whispertranscriptionconfig) configuration will be used. Defaults to None.
* <mark style="color:orange;">**input\_class**</mark>**&#x20;(Type\[**[**Input**](/core/schemas.md#input)**], optional):** Class for input validation. Defaults to [**WhisperInput**](#whisperinput).
* <mark style="color:orange;">**output\_class**</mark>**&#x20;(Type\[**[**Output**](/core/schemas.md#output)**], optional):** Class for output validation. Defaults to [**WhisperOutput**](#whisperoutput).
* <mark style="color:orange;">**name**</mark>**&#x20;(Optional\[str], optional):** Name for identification. If equals to None, class name will be used. Defaults to None.

***

***

***

## <mark style="color:green;">WhisperModelConfig</mark>

Prebuild configuration that describes default parameters for GLiNER models pipeline. Subclass of [**Config**](/core/schemas.md#config).

***

### <mark style="color:blue;">\_\_init\_\_</mark>

#### Arguments:

* <mark style="color:orange;">**name**</mark>**&#x20;(str, optional):** One of the official model names listed by whisper.available\_models(), or path to a model checkpoint containing the model dimensions and the model state\_dict. Defaults to "base".
* <mark style="color:orange;">**device**</mark>**&#x20;(Union\[str, torch.device]):** The PyTorch device to put the model into.
* <mark style="color:orange;">**download\_root**</mark>**&#x20;(Optional\[str], optional):** Path to download the model files; by default, it uses "\~/.cache/whisper".
* <mark style="color:orange;">**in\_memory**</mark>**&#x20;(bool, optional):** Whether to preload the model weights into host memory. Defaults to False.

***

***

***

## <mark style="color:green;">WhisperTranscriptionConfig</mark>

Prebuild configuration that describes default parameters for GLiNER models pipeline. Subclass of [**Config**](/core/schemas.md#config).

***

### <mark style="color:blue;">\_\_init\_\_</mark>

#### Arguments:

* <mark style="color:orange;">**verbose**</mark>**&#x20;(Optional\[bool]):** Whether to display the text being decoded to the console. If True, displays all the details, If False, displays minimal details. If None, does not display anything.
* <mark style="color:orange;">**temperature**</mark>**&#x20;(Union\[float, Tuple\[float, ...]]):** Temperature for sampling. It can be a tuple of temperatures, which will be successively used upon failures according to either compression\_ratio\_threshold or logprob\_threshold.
* <mark style="color:orange;">**compression\_ratio\_threshold**</mark>**&#x20;(float):** If the gzip compression ratio is above this value, treat as failed.
* <mark style="color:orange;">**logprob\_threshold**</mark>**&#x20;(float):** If the average log probability over sampled tokens is below this value, treat as failed.
* <mark style="color:orange;">**no\_speech\_threshold**</mark>**&#x20;(float):** If the no\_speech probability is higher than this value AND the average log probability over sampled tokens is below logprob\_threshold, consider the segment as silent.
* <mark style="color:orange;">**condition\_on\_previous\_text**</mark>**&#x20;(bool):** If True, the previous output of the model is provided as a prompt for the next window; disabling may make the text inconsistent across windows, but the model becomes less prone to getting stuck in a failure loop, such as repetition looping or timestamps going out of sync.
* <mark style="color:orange;">**word\_timestamps**</mark>**&#x20;(bool):** Extract word-level timestamps using the cross-attention pattern and dynamic time warping, and include the timestamps for each word in each segment.
* <mark style="color:orange;">**prepend\_punctuations**</mark>**&#x20;(str):** If word\_timestamps is True, merge these punctuation symbols with the next word.
* <mark style="color:orange;">**append\_punctuations**</mark>**&#x20;(str):** If word\_timestamps is True, merge these punctuation symbols with the previous word.
* <mark style="color:orange;">**decode\_options**</mark>**&#x20;(Optional\[Dict\[str, Any]]):** Keyword arguments to construct DecodingOptions instances.

***

***

***

## <mark style="color:green;">WhisperInput</mark>

Subclass of[ **IOModel**](/core/schemas.md#iomodel).

***

### <mark style="color:blue;">\_\_init\_\_</mark>

#### Arguments:

* <mark style="color:orange;">**audio**</mark>**&#x20;(NDArray\[Any]):** Audio waveform.
* <mark style="color:orange;">**initial\_prompt**</mark>**&#x20;(Optional\[str], optional):** Optional text to provide as a prompt for the first window. This can be used to provide, or "prompt-engineer" a context for transcription, e.g. custom vocabularies or proper nouns to make it more likely to predict those word correctly.

***

***

***

## <mark style="color:green;">WhisperOutput</mark>

Subclass of[ **IOModel**](/core/schemas.md#iomodel).

***

### <mark style="color:blue;">\_\_init\_\_</mark>

#### Arguments:

* <mark style="color:orange;">**text**</mark>**&#x20;(str):** Transcribed text.

***

***