Architecture · BYOK

Bring Your Own Key: professional control for AI interpretation

SpeechKey's core architecture principle

Most translation apps hide the AI stack behind a subscription. SpeechKey takes a different approach: you connect your own AI provider keys and decide which services handle speech recognition, translation, and voice output. This makes SpeechKey more transparent, more configurable, and better suited for professional environments where cost, data path, region, and provider choice matter.

01What BYOK means

BYOK means: you create accounts at AI providers, generate API keys, and enter them into SpeechKey. The app uses those keys to connect directly to the providers.

  • You create provider accounts (iFLYTEK, OpenAI, DeepSeek, MiniMax, ElevenLabs)
  • You generate API keys at each required provider
  • You enter the keys into SpeechKey Settings
  • SpeechKey uses those keys to connect directly to providers
  • SpeechKey does not resell AI minutes in the consumer BYOK model
  • You keep full control over your provider relationship
Architecture note In the consumer BYOK app, SpeechKey does not operate a server in the audio or transcript processing path. Audio goes directly from your device to the provider you configured.

02What BYOK does not mean

It is important to be precise here. BYOK is not a promise of complete offline processing or zero data transfer.

Important BYOK does not mean "nothing leaves your device." Live AI interpretation requires speech and text to be processed by the providers you configure. BYOK means SpeechKey does not insert its own server layer into the consumer audio/transcript path and does not own your provider relationship.
  • It does not mean all processing stays on your device
  • It does not mean providers cannot process or retain data
  • It does not mean there are no provider costs
  • It does not mean zero setup
  • It does not mean enterprise-grade compliance automatically
  • Provider-side logging, retention, and training rules depend on the provider account and terms you use

03BYOK benefits

Privacy by architecture

No SpeechKey account, no SpeechKey analytics, no SpeechKey transcript storage in the consumer model. SpeechKey is data-minimised by design — not by promise.

Cost transparency

You pay providers directly at their published rates. No hidden per-minute resale model. Your provider dashboard shows exactly what was consumed.

Provider independence

Switch providers anytime. No SpeechKey business model that penalises you for switching. When a better model becomes available, enable it in settings.

Regional flexibility

Use global providers for best quality or China-compatible providers for lower latency and no-VPN scenarios in mainland China.

Quality tuning

Choose the model that fits your use case. Upgrade to GPT-4o flagship for premium quality, or use cost-efficient models for frequent meetings.

Enterprise compatibility path

Teams can coordinate individual BYOK accounts. Enterprise managed services (central key management, SSO, audit logs) are available separately.

04BYOK responsibilities

With provider control comes provider responsibility. As a BYOK user, you are responsible for:

  • Provider accounts
    Creating and managing accounts at each required provider
  • Provider billing
    Ensuring active payment methods and credits at your providers
  • Provider terms
    Accepting and complying with provider terms of service and privacy policies
  • API key security
    Keeping keys secure and not sharing them. SpeechKey stores them in the iOS Keychain.
  • Usage monitoring
    Setting usage limits in provider dashboards to avoid unexpected charges
  • Pipeline selection
    Choosing the right pipeline for your use case, region, and cost tolerance

05BYOK setup paths

Depending on your use case, we recommend different configurations.

China Business

China business setup

Recommended for users in mainland China. Uses exclusively Chinese cloud providers — no VPN required.

  • STT
    iFLYTEK — Best Chinese speech recognition
  • Translation
    DeepSeek — China-cloud LLM, cost-efficient
  • TTS
    MiniMax — Premium Chinese voice, optional
  • OpenAI
    Not recommended if VPN latency is an issue
Best quality

Best quality setup

Recommended for quality-first global users. VPN required in China.

  • STT
    iFLYTEK — Chinese ASR
  • Translation
    OpenAI GPT-4o flagship
  • TTS
    MiniMax or ElevenLabs
  • Pipeline
    Pro Live
Low cost

Low-cost setup

Recommended for frequent meetings with cost sensitivity.

  • STT
    iFLYTEK
  • Translation
    OpenAI GPT-4o-mini or DeepSeek
  • TTS
    iOS system voice (free)
  • Pipeline
    Standard
Silent support

Silent meeting support

Recommended for meetings where audio output is not appropriate. Transcript bubbles on Apple Watch.

  • STT + Translation
    Active
  • TTS
    Disabled
  • Apple Watch
    Primary display
  • Pipeline
    Watch-Modus

06Consumer BYOK architecture flow

Consumer BYOK flow
Microphone → SpeechKey App → Your STT provider → Your LLM provider → Your TTS provider → Output
SpeechKey servers: not in the consumer audio/transcript path
Provider-side processing is governed by the terms and settings of the provider accounts you use.

07Decision helper

Answer these four questions to find the right setup path.

Question 1

Are you mainly using SpeechKey in China?

Question 2

Do you need spoken output or captions only?

Question 3

Is lowest latency or lowest cost more important?

Question 4

Are you comfortable creating API keys from provider dashboards?

08Get started

Typical first setup takes 10–20 minutes depending on provider accounts. After setup, switching pipelines takes only a few seconds.

View setup guide → Compare pipelines See pricing