# How to get a speech-to-text API key

> An API key lets Side Whisper send your dictation to a speech-to-text provider, on your own account. Here is where each provider Side Whisper supports keeps its keys, which ones you can start with for free, and how to add yours.

Source: https://sidewhisper.app/docs/voice-api-keys/

## What an API key is, and why use one

An API key is a long code, a little like a password, that a provider gives you so an app can use its service on your account. In Side Whisper it does one thing: with Pro, it lets a speech-to-text provider you choose write your dictation, in place of the speech model on your PC.

You do not need a key to dictate. The speech models on your PC work offline, on Free and on Pro, and [Offline voice dictation](https://sidewhisper.app/docs/offline-voice/) explains them. A key is for when they are not enough for you: the provider’s model runs on its own computers, so your PC runs no speech model while you dictate, and you pick the provider whose model suits your languages.

With your key, each dictation goes from your PC straight to that provider, under your account and its terms, never through Side Whisper. Side Whisper keeps the key on your PC, encrypted by Windows, and sends it only to that provider. Whatever the provider charges, it charges you directly: Side Whisper adds nothing per dictation.

> **Pro.** Dictating with your own API key is part of Pro. On Free, dictation stays on your PC and Settings shows the PRO badge beside the API key. [Compare plans](https://sidewhisper.app/pricing/), or see what [Voice](https://sidewhisper.app/voice/) does on each plan.

## Which providers are free to start

Every provider gives you an API key for free. What differs is whether the key can transcribe before you pay. Groq, Deepgram, AssemblyAI, ElevenLabs and Mistral let a new account dictate for free, with no card. Google Cloud gives free minutes once a card is on file. OpenAI needs prepaid credit first.

| Provider | Free to start | What you get for free | Card to start |
| --- | --- | --- | --- |
| [Groq](https://sidewhisper.app/docs/voice-api-keys/#groq) | Yes | A free plan, with limits per minute, hour and day that leave room for hours of audio a day | No |
| [Deepgram](https://sidewhisper.app/docs/voice-api-keys/#deepgram) | Yes | $200 of credit, kept until you use it | No |
| [AssemblyAI](https://sidewhisper.app/docs/voice-api-keys/#assemblyai) | Yes | $50 of credit | No |
| [ElevenLabs](https://sidewhisper.app/docs/voice-api-keys/#elevenlabs) | Yes | 10,000 credits a month on its free plan, about 30 minutes of speech to text | No |
| [Mistral](https://sidewhisper.app/docs/voice-api-keys/#mistral) | Yes | Free mode, with usage and rate limits | No |
| [Google Cloud](https://sidewhisper.app/docs/voice-api-keys/#google-cloud) | Yes, with a card on file | 60 minutes a month, and $300 of credit for 90 days for a new Google Cloud customer | Yes |
| [OpenAI](https://sidewhisper.app/docs/voice-api-keys/#openai) | No | Nothing: dictation uses prepaid credit, from $5 | Yes |
| [OpenAI-compatible](https://sidewhisper.app/docs/voice-api-keys/#openai-compatible) | Depends on the server | A server you run on your own PC needs no account | Depends on the server |

> **Tip.** To try it without paying or adding a card, start with Groq, Deepgram or AssemblyAI.

These are each provider’s own offers, as its pages show them. A provider can change them, and its page has the current terms. Credit runs out as you use it, and a free plan’s limits start again after each period. Side Whisper never charges for what you send to a provider.

## OpenAI API key

OpenAI’s API has its own billing, apart from ChatGPT: a ChatGPT subscription does not pay for it. It runs on prepaid credit, so add some before you dictate. The smallest purchase is $5.

1. Sign up or log in at [platform.openai.com](https://platform.openai.com).
2. Open [Billing](https://platform.openai.com/settings/organization/billing), add a payment method and buy credit.
3. Open [API keys](https://platform.openai.com/settings/organization/api-keys) and choose **Create new secret key**. Name it Side Whisper, keep the other choices, and create it.
4. Copy the key right away. It starts with `sk-`.

In Side Whisper, choose **OpenAI** under Provider. The model starts on `gpt-transcribe`, the most accurate; `gpt-4o-transcribe`, `gpt-4o-mini-transcribe` and `whisper-1` are in the list too. Clean up the text works with OpenAI.

Test key checks the key, not your credit: if Settings then says OpenAI is limiting this key, or the pill says Provider limit reached, add credit. To cap what you spend, set a spend limit under [Limits](https://platform.openai.com/settings/organization/limits). OpenAI’s [quickstart](https://developers.openai.com/api/docs/quickstart) has its own steps.

## Groq API key

Groq’s free plan needs no card. It limits how many requests and how much audio you send each minute, hour and day, with room for hours of audio a day, and its [rate limits page](https://console.groq.com/docs/rate-limits) lists the numbers. A paid plan, when you want more, asks for a payment method.

1. Sign up or log in at [console.groq.com](https://console.groq.com).
2. Open [API Keys](https://console.groq.com/keys) and choose **Create API Key**.
3. Give it a name, such as Side Whisper, and create it.
4. Copy the key it shows. It starts with `gsk_`.

In Side Whisper, choose **Groq** under Provider. The model starts on `whisper-large-v3`, the more accurate of the two; `whisper-large-v3-turbo` is faster. Clean up the text works with Groq.

## Deepgram API key

Every new Deepgram account gets $200 of free credit, with no card, and the credit stays until you use it, as [Deepgram’s pricing](https://deepgram.com/pricing) says.

1. Sign up at [console.deepgram.com](https://console.deepgram.com/signup).
2. In the Deepgram Console, with your project picked at the top left, open **Settings**, then **API Keys**, and choose **Create a New API Key**.
3. Name it Side Whisper, keep the permissions it offers, and choose **Create Key**.
4. Copy the key and choose **Got it**. Deepgram shows it only once. Its keys are 40 characters of 0 to 9 and a to f.

In Side Whisper, choose **Deepgram** under Provider. The model starts on `nova-3`, the most accurate; `nova-2` is in the list too. Deepgram’s [own steps](https://developers.deepgram.com/docs/create-additional-api-keys) show the same screens.

Side Whisper asks Deepgram to leave every dictation out of its [model improvement program](https://developers.deepgram.com/docs/the-deepgram-model-improvement-partnership-program), so Deepgram does not train its models on your speech and keeps it only to answer the request. This is always on, with no setting to change. Deepgram may bill such requests at a higher rate than its listed one; its usage log shows the rate.

## ElevenLabs API key

ElevenLabs puts a new account on its free plan: 10,000 credits a month. Speech to text uses 330 credits a minute, so the free plan covers about 30 minutes of dictation a month. [Its pricing](https://elevenlabs.io/pricing) lists both numbers.

1. Sign up at [elevenlabs.io](https://elevenlabs.io/app/sign-up).
2. Choose **Developers** in the left sidebar, then the **API Keys** tab, and create a key.
3. A key reaches only the features you allow: give this one access to **Speech to Text**.
4. Copy the key when it appears. ElevenLabs shows it in full only once.

In Side Whisper, choose **ElevenLabs** under Provider. The model starts on `scribe_v2`, the most accurate; `scribe_v1` is in the list too. Test key transcribes half a second of silence, so the key needs its Speech to Text access. ElevenLabs’ [own steps](https://elevenlabs.io/docs/help-center/technical/how-do-i-authorize-myself-using-an-api-key) show the same screens.

## AssemblyAI API key

AssemblyAI gives a new account $50 of free credit, with no card, as [its pricing](https://www.assemblyai.com/pricing) says.

1. Sign up at [assemblyai.com](https://www.assemblyai.com/dashboard/signup).
2. In the dashboard, open **API Keys**, under Workspace, then Manage. Copy the key that is there, or choose **Create New API Key**, name it Side Whisper and choose **Create**. Its keys are 32 characters of 0 to 9 and a to f.

In Side Whisper, choose **AssemblyAI** under Provider. The model starts on `universal-3-5-pro`, the most accurate in its 18 languages; a dictation in another language, Romanian among them, goes to `universal-2`. AssemblyAI’s [own steps](https://www.assemblyai.com/docs/account-management) show the same screens.

Under **Region**, just below Provider, choose where AssemblyAI processes and stores your audio and text: **United States**, AssemblyAI’s default, or **European Union**, at the same price. AssemblyAI keeps your account and usage data in the United States either way, as [its region page](https://www.assemblyai.com/docs/pre-recorded-audio/select-the-region) says. Test key checks the key in the region you chose.

AssemblyAI keeps each dictation in your account, in that region, until Side Whisper deletes it, once the text is back. [Dictate with your own API key](https://sidewhisper.app/docs/offline-voice/#own-api-key) says when one stays.

## Mistral API key

Mistral’s Studio starts in Free mode, with no card, and limits how much you use. In Free mode, Mistral says it may use what you send to train its models. To turn that off, open [admin.mistral.ai](https://admin.mistral.ai), choose **Privacy** and turn off the switch under **Anonymous improvement data**.

1. Sign up or log in at [console.mistral.ai](https://console.mistral.ai).
2. Choose **API Keys** in the left sidebar, then **Create new key**.
3. Give it a name, such as Side Whisper, and an expiry date if you want one, then create it.
4. Copy the key from the dialog. Mistral shows it only once.

In Side Whisper, choose **Mistral** under Provider. The model is `voxtral-mini-latest`, Voxtral Mini Transcribe, which transcribes 13 languages: English, Chinese, Hindi, Spanish, Arabic, French, Portuguese, Russian, German, Japanese, Korean, Italian and Dutch. Clean up the text works with Mistral. Mistral’s [own steps](https://docs.mistral.ai/getting-started/quickstarts/studio/activate-and-generate-api-key) show the same screens.

## Google Cloud Speech-to-Text API key

Google Cloud gives every account 60 minutes of Speech-to-Text a month, and a new Google Cloud customer $300 of credit to use within 90 days. Both need a billing account with a card or another payment method on file. Google bills nothing during those 90 days; when they end, it stops the project unless you upgrade to a paid account, which then pays for what goes past the free minutes. [Google’s free offers](https://docs.cloud.google.com/free/docs/free-cloud-features) has the details.

1. Sign in to the [Google Cloud console](https://console.cloud.google.com) with your Google account, and take the $300 offer if the console shows it.
2. Create a project, or pick one, and link a billing account when Google asks.
3. Click the search field at the top, type speech, open **Cloud Speech-to-Text API** and choose **Enable**.
4. Open the [Credentials](https://console.cloud.google.com/apis/credentials) page, choose **Create credentials**, then **API key**.
5. Add a restriction under **API restrictions**: choose **Restrict key** and select **Cloud Speech-to-Text API**. Leave the service account box unticked, then choose **Create**.
6. Copy the key from the **API key created** dialog. It starts with `AIza` and is 39 characters long.

In Side Whisper, choose **Google Cloud** under Provider. The model starts on `latest_long`, made for dictation; `latest_short` is in the list too. Google needs a language, so choose yours under Language: Detect language sends English (United States).

Test key transcribes half a second of silence, so it fails until the API is enabled and billing is linked. Google’s own steps: [turn on Speech-to-Text](https://docs.cloud.google.com/speech-to-text/docs/v1/before-you-begin) and [create an API key](https://docs.cloud.google.com/docs/authentication/api-keys).

## OpenAI-compatible server

Choose **OpenAI-compatible** for a speech-to-text server that takes OpenAI’s audio requests: one you run on your own PC, or a service on the web that offers that API. A server on your PC needs no account, and its key is optional. A service on the web sets its own terms and prices.

1. Find the server’s OpenAI-style address in its documentation, such as `http://localhost:8000/v1` for a server on this PC.
2. In Side Whisper, choose **OpenAI-compatible** under Provider and type the address under **Server address**. It must start with https://, or with http:// for a server on this PC.
3. If the server needs a key, copy it and press **Paste key**. Side Whisper keeps it for that address only.
4. Press **Test key** to read the server’s list of models, or type the model’s name under **Model name**.

Clean up the text is not offered for a compatible server: its text is typed as heard.

## Add your key to Side Whisper

1. Copy the key from the provider’s page.
2. In Side Whisper, open Settings, then **Voice**, then **Models**. **Your own API key** is under the model cards.
3. Choose your provider under **Provider**. For an OpenAI-compatible server, type its address under **Server address** first.
4. Press **Paste key**. Side Whisper reads the key from the clipboard, saves it on this PC and clears the clipboard. Windows clipboard history (`Win+V`) keeps its own copy until you delete it there.
5. Press **Test key**. The line under the key says The key works, or why the provider refused it.
6. Keep the **Model** Settings starts on, the provider’s most accurate, and set **Language** to the language you speak.
7. At the top of the page, under **Dictate with**, choose **Your API key**. The notice says where your audio goes: choose **Send audio to** your provider to agree.
8. Dictate as usual with `Alt+A`.

Your API key can be chosen once a key and a model are saved. If the provider cannot answer, the voice model you chose on this PC types the text, when it is downloaded, and the pill says Inserted (local model) or Copied (local model). [Dictate with your own API key](https://sidewhisper.app/docs/offline-voice/#own-api-key) lists what the provider receives and how Clean up the text works.

## If the key does not work

Settings, then Voice, then Models, says what went wrong under the key, with the provider’s own reason when it gives one. The dictation pill says it in a few words.

| What you see | What to do |
| --- | --- |
| “This does not look like a key from …” | What you copied may be something else, such as the key’s name. Copy the key itself and press Paste key again. Test key tells whether it works. |
| “… refused the key.” or “API key refused” | The key is incomplete, or it was deleted at the provider. Create a key again and paste it. With Google Cloud, also check that Cloud Speech-to-Text API is enabled for the key’s project, that billing is linked, and that the key’s API restrictions allow it. |
| “… is limiting this key” or “Provider limit reached” | The account is out of credit, or reached its plan’s limit for now. Add credit, or wait for the limit to start again. Test key checks the key, not your credit, so a key can pass the test and still be limited. |
| “… refused the request. Check the model.” or “Provider refused request” | Choose the model Settings started on under Model, or another one from the provider’s list. |
| “… did not answer.” or “Provider unavailable” | The provider is down or slow. Try again later. |
| “There is no network connection.” or “No network” | Check that your PC is online. |
| “Inserted (local model)” or “Copied (local model)” | The provider could not answer, so the model on your PC wrote the text. Settings, then Voice, then Models, says why. |

## Keep your key safe

- **Keep it private.** A key works like a password for your account with the provider: whoever has it can use your credit. Never post it, send it in a message or show it in a screenshot.
- **Make one key just for Side Whisper.** You can then delete it without touching your other apps.
- **It stays on your PC.** Side Whisper keeps the key encrypted by Windows for your Windows account. It never syncs, it is never in a backup or an export, and Settings shows only its last four characters. Only its provider receives it, with each dictation you send there.
- **Clear the clipboard history.** If Windows clipboard history is on, press `Win+V` and delete the key there.
- **Remove it** under Your own API key when you stop using it, and delete it at the provider too, or as soon as you think someone else has it.

| Provider | Where to delete a key |
| --- | --- |
| OpenAI | On the [API keys](https://platform.openai.com/settings/organization/api-keys) page |
| Groq | On the [API Keys](https://console.groq.com/keys) page |
| Deepgram | In the Deepgram Console, under Settings, then API Keys |
| ElevenLabs | Under Developers, then API Keys: the key’s More Actions (…) menu, then Delete |
| AssemblyAI | Under Workspace, then Manage, then API Keys: Delete. A project keeps at least one key, so create the next one first. |
| Mistral | Under API Keys in the Studio console |
| Google Cloud | On the [Credentials](https://console.cloud.google.com/apis/credentials) page. Google can restore a deleted key for 30 days. |
| OpenAI-compatible | Where the server or service keeps its keys |