Speech Presets

Speech presets let you create shortcut replacement rules for commonly used phrases. When the ASR result matches a trigger phrase, it is automatically replaced with your preset content, greatly improving repeated input efficiency.

Overview

How it works

Voice input → ASR → exact match preset trigger → replace with preset content → insert

Logic:

1. Perform speech recognition normally
2. Match the transcript against preset triggers
3. If matched (exact or case-insensitive), replace with preset content
4. Otherwise keep original transcript

Good for

Recommended

• ✅ common phrases: email, phone number, address
• ✅ canned replies: "OK", "Received", etc.
• ✅ terms: "ASR" → "Automatic Speech Recognition"
• ✅ long templates: signatures, disclaimers
• ✅ emoji combos: e.g. "haha" → "hahaha 😄"

Data & config

Key	Type	Description
speechPresetsJson	String	preset list JSON
activeSpeechPresetId	String	active preset id (reserved; currently unused)

Preset data structure

Each preset has 3 fields:

data class SpeechPreset(
    val id: String,        // UUID
    val name: String,      // trigger phrase (what you say)
    val content: String    // replacement (what gets inserted)
)

Example:

{
  "id": "550e8400-e29b-41d4-a716-446655440000",
  "name": "my email",
  "content": "example@domain.com"
}

Usage

Create a preset

1. Open Settings → Other features → Speech presets
2. Enter a trigger phrase (e.g. "my email")
3. Enter replacement content (e.g. "example@domain.com")
4. Tap "Add"

Naming tips

• Keep trigger phrases short and easy to say
• Make triggers unique to avoid conflicts
• Prefer consistent patterns like "my ", " address", etc.

Use a preset

1. Use voice input as usual
2. Speak the trigger phrase
3. The transcript is replaced automatically
4. The final text is inserted into the editor

Edit a preset

1. Open Settings → Other features → Speech presets
2. Select a preset from the dropdown
3. Modify trigger/content
4. Tap "Update" (if available)

Note

Some versions may require deleting and re-adding a preset to modify it. Follow the actual UI behavior.

Delete a preset

1. Select the preset
2. Tap "Delete" and confirm

Matching rules

Exact match first

1. Exact match: transcript equals trigger phrase exactly (including spaces and case)
2. Case-insensitive match: same content but different case

Examples

Trigger	Transcript	Match	Type
"my email"	"my email"	✅	exact
"my email"	"my email"	❌	whitespace differs
"ASR"	"asr"	✅	case-insensitive
"received"	"received it"	❌	not equal
"ok"	" ok "	✅	trimmed spaces

Practical examples

Personal info

[
  {
    "name": "my email",
    "content": "your.email@example.com"
  },
  {
    "name": "my phone",
    "content": "13800138000"
  },
  {
    "name": "my address",
    "content": "Room XX, No. XX Road, Chaoyang District, Beijing"
  }
]

Templates

[
  {
    "name": "email signature",
    "content": "Best regards,\\n\\nJohn Doe\\nSenior Engineer\\nACME Corp\\nPhone: +1-xxx\\nEmail: john@example.com"
  },
  {
    "name": "disclaimer",
    "content": "This message is for reference only and does not constitute investment advice."
  }
]

How it interacts with other features

With AI post-processing

Speech presets run before AI post-processing:

ASR → [speech preset replacement] → [AI post-processing] → insert

With ASR providers

Speech presets are provider-agnostic:

• ✅ works for all providers
• ✅ works for cloud and local engines
• ✅ works for both streaming and non-streaming modes

Best practice: pick an accurate ASR provider so trigger phrases are recognized correctly.

With floating ball

Floating ball voice input fully supports speech presets:

Floating ball recording → ASR → preset match → insert into active editor

Notes

Avoid conflicts

• ❌ avoid very common phrases (e.g. "ok", "thanks")
• ❌ avoid too-short triggers (single syllable)
• ❌ avoid conflicting replacements for common expressions

Recommended trigger design

• ✅ use fixed patterns ("my ", " address")
• ✅ use proper nouns (company email, home address)
• ✅ use abbreviations (ASR, LLM)
• ✅ use unique phrases ("insert signature", "append disclaimer")

Performance

• Count: keep under ~50 presets (too many slows matching)
• Complexity: linear scan, O(n)
• Content length: unlimited, but very long content may affect UX