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

# Audio & Transcription

> Pawa AI APIs provide seamless audio `input and output capabilities` — `Speech-to-Text (STT)` and `Text-to-Speech (TTS)`. These enable models to chat, speak, and understand audio directly, supporting use cases like voice assistants, transcription, and audio-enhanced knowledge bases.

> **Text-to-Speech (TTS) & Speech-to-Text (STT) Examples with Pawa AI**

<img src="https://mintcdn.com/sartifycoltd/h3QvoufTX15l3Mnz/images/speech%20to%20text%20-%20text%20to%20speech.png?fit=max&auto=format&n=h3QvoufTX15l3Mnz&q=85&s=166d645de6e89a181954222916578577" alt="Audio & Transcription with Pawa AI" className="rounded-xl shadow-lg w-full" width="1932" height="1216" data-path="images/speech to text - text to speech.png" />

***

### Text-to-Speech (TTS)

**Text-to-Speech (TTS)** allows you to generate high-quality, natural-sounding speech directly from text.\
This capability is essential for building **voice-enabled applications**, making content accessible to wider audiences, and creating **immersive user experiences**.

<Note>
  By default,  our `tts api endpoint` implements streaming, so in your app you can directly use Server Side Event to get streaming of the audio back. If you dont use the SSE then the api will fallback to normal non sctreaming to wait for the full audio to be generated back give back answer.
</Note>

You can learn about streaming [here](/capabilities/streaming).

### Supported Languages For Text to Speech

* `Swahili`
* `English`

#### Use Cases

* **Voice assistants**: Let your chatbot respond with speech instead of text only.
* **Learning platforms**: Automatically generate audio versions of documents, lessons, or Q\&A sessions.
* **Accessibility tools**: Help users with visual impairments interact with your app through audio.
* **Media & podcasts**: Generate narrations from written articles or blogs.

#### Voice options

The TTS endpoint provides 3 built‑in local voices to control how speech is rendered from text. This includes

* `Ame`
* `Liora`
* `Ayana`

#### Models with Text to speech Audio Capabilities

* Pawa Text To Speech (`pawa-tts-v1-20250704`) with text to speech conversation.

Example Playback Speech
<audio controls className="mt-4 w-full"> <source src="https://mintcdn.com/sartifycoltd/h3QvoufTX15l3Mnz/images/tanzania_info2.wav?fit=max&auto=format&n=h3QvoufTX15l3Mnz&q=85&s=b4e891eff8f290e6e709459bd643fc90" type="audio/mpeg" data-path="images/tanzania_info2.wav" /> Your browser does not support the audio element. </audio>

> <p className="mt-2"> Original Text: <span className="font-medium">"Jina la jamhuri ya muungano wa Tanzania, ni nchi iliyopo Afrika ya Mashariki ndani ya ukanda wa maziwa makuu ya Afrika, imepakana na Uganda na Kenya upande wa kaskazini, Bahari ya Hindi upande wa mashariki, Msumbiji malawi na Zambia upande wa kusini, Congo, Burundi na Rwanda upande wa magharibi, eneo la Tanzania ni takribani kilometa za mraba 940 mb/h. Saa arobaini na dakika elfu 300, eneo linalokaliwa na maji ne asalimia 6.2 - Mlima Kilimanjaro - Mlima mrefu zaidi barani Afrika upo kaskazini mashariki wa Tanzania."</span> </p>

#### Text to Speech Request Example

```bash theme={null}
curl --request POST \
     --url https://api.pawa-ai.com/v1/voice/text-to-speech \
     --header "Authorization: Bearer $PAWA_AI_API_KEY" \
     --header 'Content-Type: application/json' \
      --data '{
                "model": "pawa-tts-v1-20250704",
                "voice": "ame",
                "max_tokens": 65536,
                "temperature": 0.5,
                "top_p": 0.95,
                "text": "Hello, welcome to Pawa AI. Upgrade now to enjoy Unlimited access to advanced AI"
                "repetition_penalty": 1.1
}' \
     --output speech.mp3
```

Check the audio file saved in current directory, open it to play and listen to generated audio.

### Speech-to-Text (STT)

**Speech-to-Text (STT)** converts audio into text with high accuracy. With optional `speaker identity`, `timestamps`.\
This is powerful for **transcription, audio search, summarization, and voice-enabled interfaces**.

### Supported Languages For Text to Speech

* `Swahili`
* `English`
* `Luo`
* `Meru`
* `kamba`
* `Hausa`
* `Igbo`
* `Yoruba`
* `Zulu`
* `Tswana`
* `Nyakole` etc...

#### Use Cases

* **Meeting & call centers transcription**: Turn long discussions into structured notes.
* **Customer service**: Convert call center conversations into searchable text.
* **Education**: Transcribe lectures, podcasts, and webinars.
* **Productivity**: Voice notes and dictation apps.

#### Models with Speech to Text Audio Capabilities

* Pawa Speech To Text (`pawa-stt-v1-20240701`) with audio input to text conversation.

Original Playback Speech
<audio controls className="mt-4 w-full"> <source src="https://mintcdn.com/sartifycoltd/h3QvoufTX15l3Mnz/images/tanzania_info2.wav?fit=max&auto=format&n=h3QvoufTX15l3Mnz&q=85&s=b4e891eff8f290e6e709459bd643fc90" type="audio/mpeg" data-path="images/tanzania_info2.wav" /> Your browser does not support the audio element. </audio>

> <p className="mt-2"> Example Text: <span className="font-medium">"Jina la jamhuri ya muungano wa Tanzania, ni nchi iliyopo Afrika ya Mashariki ndani ya ukanda wa maziwa makuu ya Afrika, imepakana na Uganda na Kenya upande wa kaskazini, Bahari ya Hindi upande wa mashariki, Msumbiji malawi na Zambia upande wa kusini, Congo, Burundi na Rwanda upande wa magharibi, eneo la Tanzania ni takribani kilometa za mraba 940 mb/h. Saa arobaini na dakika elfu 300, eneo linalokaliwa na maji ne asalimia 6.2 - Mlima Kilimanjaro - Mlima mrefu zaidi barani Afrika upo kaskazini mashariki wa Tanzania."</span> </p>

#### Speech to Text Request `without Speaker Diarization` Example

```bash theme={null}
curl  --request POST \
      --url https://api.pawa-ai.com/v1/voice/speech-to-text \
      --header "Authorization: Bearer $PAWA_AI_API_KEY" \
      --header 'Content-Type: multipart/form-data' \
      --form files=@sartify_info2.wav \
      --form model=pawa-stt-v1-20240701 \
      --form language=English \
      --form is_speaker_diarization=false 
```

#### Example Response Without Speaker Diarization

```json theme={null}
{
  "success": true,
  "message": "Audio transcribed succesfully",
 "data": {
    "transcriptions": [
      {
        "filename": "hello.wav",
        "transcript": "okay let's break down the products certify has built so far here's a list based on available information hasto astory poll this is the flagship product it's an ai power document processing platform essentially it uses ai to understand and process documents likely automating tasks like data extraction classification and more astor astory to warried an ai powered educational tool designed to improve learning outcomes it likely provides personalized learning experiences tutoring assistance and other educational supports as says the power models these aren't a product per se but rather the underlying ai models that power their other products there are a suite of smaller specialized ai models focusing on african languages vision and audio think of them as the engine behind docipro and tutorai it's worth noting that satilfy is a relatively young company so their product line is still"
      }
    ]
  }
}

```

Change `is_speaker_diarization` to `true` to enable speaker identity or diarization capability on the response

#### Example Response With Speaker Diarization

```json theme={null}
{
  "success": true,
  "message": "Audio transcribed succesfully",
 "data": {
    "transcriptions": [
      {
        "filename": "hello.wav",
        "transcript": [
          {
            "Start": 0,
            "End": 14.87,
            "Speaker": 0,
            "Content": "Okay, let's break down the products Certify has built so far. Here's a list based on available information. Astor Astoripo, this is their flagship product. It's an AI-powered document processing platform."
          },
          {
            "Start": 14.9,
            "End": 24.61,
            "Speaker": 0,
            "Content": "Essentially, it uses AI to understand and process documents, likely automating tasks like data extraction, classification, and more."
          },
          {
            "Start": 24.87,
            "End": 38.81,
            "Speaker": 0,
            "Content": "As well as Certifiturai, an AI-powered educational tool designed to improve learning outcomes. It likely provides personalized learning experiences, tutoring assistance, and other educational sup-"
          },
          {
            "Start": 39.05,
            "End": 47.55,
            "Speaker": 0,
            "Content": "As such, as we power models, we aren't a product per se, but rather the underlying AI models that power the other products."
          },
          {
            "Start": 47.67,
            "End": 65.76,
            "Speaker": 0,
            "Content": "There are a suite of smaller specialized AI models focusing on African languages, vision, and audio. Think of them as the engine behind Docipro and TutorAI. It's worth noting that Certify is a relatively young company, so their product line is still evolving."
          }
        ]
      }
    ]
  }
}

```

You can upload up to 10 files of types `audio/mp3`, `audio/mpeg`, audio/x-mp3, `audio/wav`, `audio/wave`, `audio/x-wav`, `audio/x-pn-wav`, `audio/aac`, `audio/m4a`, `audio/x-m4a`, `audio/x-mp4`, `audio/ogg`, `audio/opus`, `audio/x-ms-wma`, `audio/wma` with each of less `500MB`. The server runs in batch mode of 5 files default.