EdgeTTS typesafe voices and profiles

name: EdgeTTS typesafe voices overview: Make Edge TTS maximally typesafe with a curated EdgeVoice enum, prosody value types, built-in narrator profile defaults matching books tools (Ava / −4%), nested voice-map YAML parity, and a BooksWriterStudio voice ComboBox. todos:

• id: edgetts-types

content: Add EdgeVoice enum, Prosody*, EdgeVoiceCatalog/Entry, EdgeVoiceProfile(s); retarget EdgeTtsSynthesisOptions + Client defaults to Ava/-4 status: completed

• id: manuscript-settings

content: Retype ManuscriptVoiceSettings + ToEdgeTtsOptions/ToSpeechOptions; FromProfile(Narrator) status: completed

• id: voicemap-yaml

content: Rewrite VoiceMapStore for nested narrator/pauses/generation YAML; fail closed on unknown short names status: completed

• id: studio-ui

content: "BooksWriterStudio Voice tab: ComboBox + int prosody; optional profile picker; load/save via store" status: completed

• id: tests-docs

content: Update unit tests + READMEs; nested YAML fixture; verify-nuget-only status: completed isProject: false

Defaults (clean break)

Match `D:\repos\books\tools\audio\voice-map.yaml`:

• Voice: en-US-AvaNeural → EdgeVoice.EnUsAva
• Rate: -4, Pitch: 0, Volume: 0
• Scene break: 1200 ms; max chunk: 2800 chars
• Replace EdgeTtsConstants.DefaultVoice / EmmaMultilingual as the package default

Architecture

flowchart LR
  UI[BooksWriterStudio ComboBox]
  Catalog[EdgeVoiceCatalog]
  Settings[ManuscriptVoiceSettings]
  Store[VoiceMapStore]
  Opts[EdgeTtsSynthesisOptions]
  Client[EdgeTtsClient SSML]

  UI --> Catalog
  UI --> Settings
  Store --> Settings
  Settings --> Opts
  Catalog --> Opts
  Opts --> Client

• Closed set at compile time: EdgeVoice enum + catalog metadata for dropdowns
• Open set only at the wire: ShortName / SSML full name formatted inside EdgeTtsClient
• Keep existing remote DTO `EdgeVoiceInfo` for ListVoicesAsync; curated rows are a separate type EdgeVoiceEntry (same pattern as KokoroVoiceEntry)

1. EdgeTts library types

In `Novolis.Audio.Voice.EdgeTts`:

Change `EdgeTtsSynthesisOptions` to typed properties only (no public string Voice/Rate/Pitch/Volume). Client formats via catalog + ToSsml() when building SSML; drop regex ValidateProsody for the typed path.

Update `EdgeTtsClient`, `README.md`, and unit tests.

2. Manuscript settings

Update `ManuscriptVoiceSettings`:

• EdgeVoice Voice (default EnUsAva)
• ProsodyPercent / ProsodyHertz for rate/volume/pitch (defaults −4 / 0 / 0)
• Add MaxChunkChars (default 2800); map into ToSpeechOptions() if planner already uses it
• ToEdgeTtsOptions() maps typed fields
• Factory: ManuscriptVoiceSettings.FromProfile(EdgeVoiceProfiles.Narrator) (+ pronunciation overlay)

3. VoiceMapStore — books nested YAML (required for clean break)

Today `VoiceMapStore` expects flat camelCase and ignores narrator: / pauses: / generation: — loading books YAML keeps Emma/+0% defaults.

Rewrite DTO to nested shape matching books tools:

narrator:
  voice: en-US-AvaNeural   # short name in file
  rate: "-4%"
  pitch: "+0Hz"
  volume: "+0%"
pauses:
  scene_break_ms: 1200
generation:
  max_chunk_chars: 2800
pronunciation:
  Ixa: "Ick-sah"

• Load: parse nested sections; EdgeVoiceCatalog.TryParse short name → enum (unknown voice → throw clear error; no silent Emma fallback)
• Save: write nested books-compatible YAML; preserve pronunciation map; use snake_case keys books already uses
• Round-trip test against a fixture copied from books voice-map.yaml structure

4. BooksWriterStudio UI

In `MainWindow.cs` Voice tab:

• Replace free-text _voiceId with ComboBox bound to EdgeVoiceCatalog.All (DisplayMemberBinding / DisplayName)
• Rate / pitch / volume: integer fields (NumericUpDown or TextBox + int parse), same UX as books-writer sliders conceptually — not "+-4%" strings
• Optional small profile ComboBox seeded with EdgeVoiceProfiles.All (selecting Narrator resets knobs); working copy remains ManuscriptVoiceSettings
• Load/save through updated VoiceMapStore only

5. Tests and docs

• `EdgeTtsClientTests`: Ava normalize; typed bad-rate impossible — test ProsodyPercent.ToSsml / catalog parse instead
• `ManuscriptAudiobookTests`: typed settings + nested YAML round-trip; fixture with Ava/−4
• Update Manuscript / EdgeTts READMEs (drop Emma examples; document nested voice-map)

Out of scope

• Enumerating the full remote Edge catalog
• Character cast / multi-speaker maps
• Changing generate-audiobook.cs in D:\repos\books (YAML stay compatible so both tools can share the file)

Done check

• dotnet test on Novolis.Audio.Unit green
• Loading books-shaped YAML yields Ava / −4 / pronunciation
• Studio Voice tab is ComboBox + ints; save writes nested YAML
• pwsh -File novolis-governance/scripts/verify-nuget-only.ps1 exit 0 (no local feeds)