Sber SaluteSpeech

About

Sber SaluteSpeech adds a macOS cloud transcription provider backed by Sber’s SaluteSpeech REST API. It is intended for users who already have, or can create, a SaluteSpeech project in Sber Developer Studio and provide their own SaluteSpeech Authorization Key.

TypeWhisper does not include a shared Sber credential. Audio is sent to SaluteSpeech for recognition, and TypeWhisper stores the Authorization Key through its plugin secret storage.

The published macOS add-on is version 0.1.1 and requires TypeWhisper 1.4.0 or newer.

Sources: Sber SaluteSpeech REST API, Sber access-token documentation, macOS plugin PR, TypeWhisper plugin registry

Requirements & Privacy

Requirement	Details	Why it matters
Platform	macOS 14.0 or newer.	The initial TypeWhisper integration is a macOS plugin.
TypeWhisper host	1.4.0 or newer.	This is the minHostVersion in the plugin manifest.
SDK compatibility	v1.	The release workflow publishes compatible metadata to plugins-v1.json.
Credential	A SaluteSpeech Authorization Key from Sber Developer Studio.	The plugin exchanges this key for short-lived OAuth access tokens before each recognition request.
Network use	Audio is uploaded to smartspeech.sber.ru; OAuth tokens are requested from Sber’s OAuth endpoint.	This is a cloud provider, not local transcription.
Credential ownership	User supplied only.	TypeWhisper does not proxy requests through a TypeWhisper-owned Sber account.

Sources: Sber personal-account setup, Sber corporate-account setup, Sber API authentication

Get a SaluteSpeech Authorization Key

For a personal account:

1. Register or sign in to Sber Developer Studio with Sber ID.
2. Create a SaluteSpeech API project under voice technologies.
3. Choose a package or use the available Freemium allocation if it is available for your account.
4. Open the project, go to API settings, and generate the Authorization Key.
5. Copy the key immediately. Sber shows it once; generate a new key if it is lost or compromised.
6. In TypeWhisper, open Settings > Integrations > Sber SaluteSpeech and paste the Authorization Key.

For a corporate account, create or use a corporate Studio space, complete the contract or billing setup Sber requires for your organization, then generate the Authorization Key from the SaluteSpeech API project’s settings.

Sources: Sber personal-account setup, Sber corporate-account setup, Sber Authorization Key instructions

OAuth Scopes

The plugin supports the scopes exposed in its settings:

Scope	Use it for
SALUTE_SPEECH_PERS	Personal SaluteSpeech projects in a personal Studio space. This is the default.
SALUTE_SPEECH_CORP	Corporate SaluteSpeech projects on postpaid corporate terms.

Sber documents additional scope values for other account types, including prepaid corporate and legacy values. This plugin currently exposes only the two scopes above. If support for a newer SDK, scope, or account model is added before public release, update the plugin manifest and this page in the same release.

Sources: Sber API authentication, Sber personal-account scope notes, Sber corporate-account scope notes

Supported Recognition Languages

The plugin exposes Russian and English recognition in TypeWhisper.

TypeWhisper selection	SaluteSpeech language sent
Auto-detect or no language	ru-RU
ru, ru-RU, or another ru-* locale	ru-RU
en, en-US, or another en-* locale	en-US

Russian is the default because SaluteSpeech is Russian-first and because the plugin should behave predictably when TypeWhisper has no spoken-language selection. Translation is not supported.

Short and Long Audio Behavior

The plugin always sends 16 kHz mono PCM_S16LE audio.

Recording	TypeWhisper behavior	SaluteSpeech flow
Up to 2 MB of PCM data and up to 60 seconds	Sends one synchronous recognition request.	POST /rest/v1/speech:recognize with model=general and the selected language.
Larger than 2 MB or longer than 60 seconds	Uses the asynchronous workflow.	Upload audio, create a recognition task, poll task status, then download the JSON result file.

Sber documents synchronous HTTP recognition for short audio with a 2 MB and one-minute limit. Their async HTTP flow is designed for longer recorded audio and uses upload, task creation, status polling, and result download.

Sources: Sber synchronous recognition, Sber async REST flow, Sber audio formats and limits

Configuration

• Authorization Key - Paste the SaluteSpeech Authorization Key from Sber Developer Studio. TypeWhisper normalizes keys pasted with or without the Basic prefix.
• Scope - Choose SALUTE_SPEECH_PERS for personal projects or SALUTE_SPEECH_CORP for supported corporate projects.
• Model - The initial plugin uses SaluteSpeech general.
• Usage estimate - The plugin tracks successful recognition duration locally to help compare against your SaluteSpeech allocation. It is an estimate; check Sber Studio for billing truth.
• Balance correction - Optionally enter the remaining recognition minutes from Sber Studio so TypeWhisper can estimate local remaining time.

Practical Testing Notes

• Validate the Authorization Key in plugin settings before selecting the engine for production dictation.
• Test both a short clip under one minute and a longer clip over one minute. They exercise different SaluteSpeech REST paths.
• Test Russian and English explicitly from a TypeWhisper profile or workflow. Auto/no-language currently defaults to Russian.
• The plugin is cloud-only and sends audio to SaluteSpeech; use a local engine when audio must stay on device.
• Certificate or trust-store issues can make validation fail in some environments. Install the certificates required by Sber for your macOS environment if Sber’s endpoint is not trusted.
• Rate limits, account package limits, and parallel request limits are controlled by Sber. Check Sber Studio for current usage and billing status.