Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.siliconflow.cn/llms.txt

Use this file to discover all available pages before exploring further.

BiBi Keyboard is an Android intelligent voice input tool built with Kotlin. It aims to make voice input more natural and efficient through LLM and ASR technologies, and provides advanced features such as AI post-processing, floating ball input, and Little Penguin integration.

​
Large Model Configuration

Currently, new users can directly choose to use the free model APIs provided by SiliconFlow on the onboarding page:
Next, we will introduce how to further configure the APIs provided by SiliconFlow for speech recognition and AI post-processing in BiBi Keyboard.

​
Configuring the Speech Recognition API

On the app’s home page, you can enter the Speech Recognition Settings and AI Post-processing Settings from two separate entry points.
After entering the speech recognition settings, you can choose and switch the speech recognition provider. Under the SiliconFlow channel, you can choose whether to use the free service. The free service provides two ASR models:
  • TeleAI/TeleSpeechASR
  • FunAudioLLM/SenseVoiceSmall
After disabling the free service, you can use your own API Key and access more models:
  • Qwen/Qwen3-Omni-30B-A3B-Instruct (multimodal model, supports audio input, faster and better)
  • Qwen/Qwen3-Omni-30B-A3B-Thinking (multimodal model, supports audio input, best quality but slightly slower)
  • TeleAI/TeleSpeechASR
  • FunAudioLLM/SenseVoiceSmall
##Configuring the Post-processing API The post-processing feature helps optimize speech recognition results or enables specific features, such as translating after recognition or summarizing to-do items from the recognized content. SiliconFlow provides two fast models for free:
  • Qwen/Qwen-3-8B (reasoning model)
  • THUDM/GLM-4-9B-0414
After disabling the free service, you can configure your own API Key to use more models:
In addition to the built-in supported models, you can also enter other model IDs through the custom option.
Some models support a deep thinking mode toggle, allowing users to choose between faster response speed and better processing quality.

​
Usage Example

After completing the configuration, let’s test whether speech recognition is working properly:
  1. Open a text input field
  2. Perform voice input
  • Make sure the current input method is Say Something
  • Press and hold the microphone button on the keyboard (the large button) to start speaking
  • Release the button after speaking and wait for the recognition result
  1. Check the result
  • If configured correctly, the recognition result will be automatically entered into the text field
  • If an error occurs, the error message will be automatically copied to the clipboard. Please check:
    • Whether the API Key is correct
    • Whether the network connection is working properly
    • Whether microphone permission has been granted
    • Whether there is audio input (check the volume waveform)
For more detailed usage instructions, see the official BiBi Keyboard documentation.