What is Deepgram? Turkish speech to text, pricing, free credit and use in voice agents

You send Deepgram an audio file or a live phone conversation through its API and get back the transcribed text (STT); it also has text-to-speech and voice agent APIs. Turkish transcription is available in Nova-3, its most accurate general model, and works in live mode too; Flux, the model built for voice agents, and Deepgram's text-to-speech don't support Turkish. Sign-up gives you $200 in free credit with no card required; after that you pay per minute. By default, audio is retained to improve the models; to turn that off, you add a parameter to every request.

What it does: Speech-to-text API
Free: $200 credit, no card
Turkish transcription: In Nova-3, live too
Turkish text to speech: No
Data: Retained by default; can be turned off
Price: Live $0.0048/min · file $0.0043/min

What does Deepgram do?

Deepgram is an AI API: it adds speech-to-text (STT) to your program or automation. You send an audio recording and get its text back, or you stream a phone conversation live and the text arrives instantly. The same company also offers text-to-speech (TTS) and a ready-made Voice Agent API.

Transcribe a recording into Turkish textWhat you get: Text (JSON)Needs a separate setup Transcribe live speech in real timeWhat you get: Real-time textNeeds a separate setup Transcription provider in a voice agentWhat you get: The agent understands the callerNeeds a connection

Who it's for: people building a voice agent that answers customers on the phone, businesses that want to turn call recordings or voicemails into text, and the developers who set these up. The no-code way to use it is to pick Deepgram as the provider in a voice agent platform that supports it.

How do you get started with Deepgram? API key and first request

Create an account Sign up at console.deepgram.com/signup; $200 in credit is added.
Create an API key Generate an API key in the console and note it down. An API key works like a secret password your program uses to prove its identity to Deepgram.
Put the key in an environment variable On the command line, type export DEEPGRAM_API_KEY="your-key"; don't write the key into code or a shared file.
Send a Turkish recording The command below is the official documentation's example, set for Turkish and with data retention turned off.

Transcribe a Turkish audio file (curl)

› curl --request POST \
  --header "Authorization: Token $DEEPGRAM_API_KEY" \
  --header 'Content-Type: audio/wav' \
  --data-binary @recording.wav \
  --url 'https://api.deepgram.com/v1/listen?model=nova-3&language=tr&smart_format=true&mip_opt_out=true'

The response comes back as JSON; the text is in the results › channels › alternatives › transcript field. smart_format=true formats dates, amounts and phone numbers for readability. Official SDKs: @deepgram/sdk on npm and deepgram-sdk on PyPI; both can read the key from the DEEPGRAM_API_KEY environment variable.

How do you use Deepgram in a voice agent?

A phone voice agent has three parts: STT that transcribes the caller, a language model that works out the reply, and TTS that speaks the reply. Deepgram handles the first part. The easiest way is to pick Deepgram as the transcription provider in the voice agent platform: Retell has it in the speech recognition settings, Vapi in the assistant's transcriber setting. Set the language to Turkish; costs and options are listed on that platform's page.

Guide to setting up a voice agent on your Turkish number with Retell

Guide to setting up a voice agent with Vapi

Deepgram's own Voice Agent API has no Turkish voice for text to speech; for an agent that speaks Turkish, you need a different provider for the voice. Flux, the model built for voice agents, doesn't cover Turkish either; for Turkish, Nova-3 is used.

Is Deepgram free? Pricing and free credit

Sign-up is free and gives you $200 in credit; no card is required. When the credit runs out, Pay As You Go continues: there is no minimum. Prices are in US dollars; the page doesn't say how taxes apply.

Item	Pay As You Go	Note
Live transcription, single language (Nova-3)	$0.0048/min	Limited-time promotion; normally $0.0077
Live transcription, multilingual	$0.0058/min	Normally $0.0092
File transcription, single language	$0.0043/min	Recordings, voicemail
Speaker diarization	Included for files	$0.0020/min live
Redaction of personal data	$0.0020/min	Removes details such as card and phone numbers
Voice Agent API, Standard	$0.075/min	Based on connection time

The Growth plan gives up to 20 percent off with prepaid annual credit of $4,000 or more. Pay As You Go allows 150 concurrent live requests and 50 file requests; the limit is per project.

Official pricing page last checked on October 28, 2026.

Does Deepgram support Turkish?

Turkish transcription (Nova-3)Yes	Language code `tr` or `tr-TR`. The documentation recommends this model for both files and live (streaming) use.
Turkish live transcriptionYes	The `language` parameter works in live mode with Nova models.
Voice agent model (Flux)No	This model, which detects turn-taking on its own, covers English and 10 languages; Turkish is not on the list.
Multilingual mixed speech (multi)No	The mode for switching languages within one conversation doesn't cover Turkish.
Turkish text to speech (TTS)No	The text-to-speech models cover English, Spanish, German, French, Dutch, Italian and Japanese.
Turkish interface and local-currency priceNo	The site, console and prices are in English and US dollars.

If you don't specify a language, Deepgram assumes English; if you don't specify a model, the older base model runs. For Turkish, add model=nova-3&language=tr to the request. The official sources don't publish a Turkish accuracy rate; test with your own recordings.

Does Deepgram store my audio or use it for training?

Default: retained

Requests are included in the Model Improvement Program (MIP) by default: audio, transcripts and synthesized speech are retained to improve the models.

Turning it off

Add mip_opt_out=true to every request (in the Settings message for Voice Agent). For these requests, audio and text are not kept after the response; request metadata and usage logs stay for 90 days. For an account-wide setting, you talk to the sales team.

Where it is processed

The default endpoint is global; the privacy policy mentions US servers. For processing inside the EU, use api.eu.deepgram.com; a full in-region guarantee also requires mip_opt_out=true.

KVKK

Deepgram lists SOC 2, GDPR and HIPAA (enterprise); it doesn't mention KVKK (Turkey's Personal Data Protection Law). A customer's voice on a phone call is personal data and is transferred abroad: get a legal review for your privacy notice and the cross-border transfer. This is not legal advice.

Frequently asked questions

What is Deepgram?

A speech-to-text API. It transcribes audio files or live phone conversations, and also offers text-to-speech and voice agent APIs.

Does Deepgram support Turkish?

For transcription, yes: in the Nova-3 model with language code tr, in live mode too. Flux, the voice-agent model, and Deepgram text to speech don't support Turkish.

Is Deepgram free?

Sign-up gives you 200 dollars in credit and no card is required. When the credit runs out, you pay per minute.

How much does Deepgram cost?

On Pay As You Go, live transcription is 0.0048 dollars per minute (promotion) and file transcription 0.0043 dollars; the Voice Agent API is 0.075 dollars per minute. Prices are in US dollars.

How do I get a Deepgram API key?

Create a free account at console.deepgram.com and generate an API key in the console. Put the key in an environment variable, not in your code.

Does Deepgram store my audio recordings?

By default it retains them for model improvement. If you add mip_opt_out=true to every request, audio and text are not kept after the response.

Related guides

Vapi

Make

Retell AI

Azure AI Speech EMA Lightning

Gemini TTS

Sources

Not clear in the official sources: Turkish accuracy rate · a separate expiry date for the $200 credit · card payment and access restrictions from Turkey · whether taxes are added to prices · any KVKK-specific statement. Each one is covered in its section along with what you can do.