> ## Documentation Index
> Fetch the complete documentation index at: https://docs.getpatter.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Whisper STT (OpenAI)

> OpenAI Whisper REST and GPT-4o transcribe family for Patter pipeline mode.

# Whisper STT

`WhisperSTT` is a buffered HTTP transcription adapter for OpenAI's [`POST /v1/audio/transcriptions`](https://platform.openai.com/docs/api-reference/audio/createTranscription) endpoint. It buffers \~1 s of incoming PCM audio (16 kHz, 16-bit mono), wraps it as a WAV blob, and submits it to Whisper for transcription. Drop-in compatible with the streaming `STTProvider` interface so it can be swapped for Deepgram / Soniox / Speechmatics without changes to the calling code.

For \~10x lower latency see the GPT-4o transcribe family below — it's a strict subclass that hits the same endpoint with `gpt-4o-transcribe` / `gpt-4o-mini-transcribe`.

## Install

`whisper` ships in the base install — no extra needed.

```bash theme={null}
pip install getpatter
```

## Usage

<Note>
  `getpatter.stt.whisper.STT()` and `getpatter.stt.openai_transcribe.STT()`
  both auto-resolve `OPENAI_API_KEY` from the environment when `api_key=`
  is omitted.
</Note>

<CodeGroup>
  ```python Python theme={null}
  # Whisper-1 (REST)
  from getpatter.stt import whisper

  stt = whisper.STT()                                       # reads OPENAI_API_KEY
  stt = whisper.STT(api_key="sk-...", language="it")

  # GPT-4o transcribe family — ~10x faster than whisper-1
  from getpatter.stt import openai_transcribe

  stt = openai_transcribe.STT(model="gpt-4o-transcribe")    # default
  stt = openai_transcribe.STT(model="gpt-4o-mini-transcribe")
  ```

  ```typescript TypeScript theme={null}
  // See typescript-sdk/providers/whisper for the TS twin.
  ```
</CodeGroup>

Plug it into an agent:

```python theme={null}
import asyncio
from getpatter import Patter, Twilio, ElevenLabsTTS
from getpatter.stt import openai_transcribe

phone = Patter(carrier=Twilio(), phone_number="+15550001234")

agent = phone.agent(
    stt=openai_transcribe.STT(),                          # OPENAI_API_KEY from env
    tts=ElevenLabsTTS(voice_id="rachel"),
    system_prompt="You are a helpful assistant.",
)

asyncio.run(phone.serve(agent))
```

## Models and rates

Per minute of audio (defaults from `getpatter.pricing`):

| Provider key        | Model                         | Rate / min |
| ------------------- | ----------------------------- | ---------- |
| `whisper`           | `whisper-1` (default)         | \$0.006    |
| `whisper`           | `gpt-4o-transcribe`           | \$0.006    |
| `whisper`           | `gpt-4o-mini-transcribe`      | \$0.003    |
| `openai_transcribe` | `gpt-4o-transcribe` (default) | \$0.006    |
| `openai_transcribe` | `gpt-4o-mini-transcribe`      | \$0.003    |
| `openai_transcribe` | `whisper-1`                   | \$0.006    |

The two provider keys hit the same endpoint but are tracked separately in the dashboard so cost attribution stays clean.

## Languages

`language="en"` by default. Whisper-1 and the GPT-4o transcribe family auto-detect the spoken language but accept an explicit BCP-47 hint (e.g. `"it"`, `"fr"`, `"es"`, `"de"`, `"pt"`, `"ja"`, `"zh"`) for higher accuracy on short utterances. See the [OpenAI language coverage list](https://platform.openai.com/docs/guides/speech-to-text/supported-languages).

## Options

| Option            | Default                                                      | Notes                                                                                 |
| ----------------- | ------------------------------------------------------------ | ------------------------------------------------------------------------------------- |
| `api_key`         | `None`                                                       | Reads from `OPENAI_API_KEY` when omitted.                                             |
| `model`           | `"whisper-1"` (Whisper) / `"gpt-4o-transcribe"` (Transcribe) | Restricted to the family's allowed model set; misconfigured calls raise `ValueError`. |
| `language`        | `"en"`                                                       | BCP-47 code.                                                                          |
| `response_format` | `"json"`                                                     | Pass `"verbose_json"` to surface per-segment confidence and timestamps.               |

## Low-level usage

```python theme={null}
from getpatter.providers.whisper_stt import WhisperSTT
from getpatter.providers.openai_transcribe_stt import OpenAITranscribeSTT

stt = WhisperSTT(api_key="sk-...", model="whisper-1", language="en")
fast = OpenAITranscribeSTT(api_key="sk-...", model="gpt-4o-mini-transcribe")

await stt.connect()
await stt.send_audio(pcm_chunk)                           # 16 kHz PCM s16le
async for t in stt.receive_transcripts():
    print(t.text, t.is_final, t.confidence)
await stt.close()
```