feature: elevenlabs tts provider

[ilteoood]

Linked issue

Closes #2008

Type of change

• Bug fix
• New feature (with prior issue + design alignment)
• Refactor / code improvement
• Docs

What does this PR do

Supports ElevenLabs as a tts provider

How did you verify it

Checklist

• One focused change. Unrelated cleanups split out.
• npm run typecheck passes
• npm run lint passes
• npm run format ran (Biome)
• UI changes include screenshots or video for every affected platform
• Tests added or updated where it made sense

[greptile-apps]

Greptile Summary

Adds ElevenLabs as a third TTS provider alongside OpenAI and local (Sherpa), routed through the existing speech-service pipeline. The feature is well-scoped: it lives in a dedicated elevenlabs/ directory, integrates through the same local→openai→elevenlabs initialization chain used by other providers, and previous review feedback on reader-lock cleanup, the identity-comparison label function, the vi.hoisted/vi.stubGlobal test pattern, and the dead null guard have all been addressed.

• TextToSpeechProvider gains a required readonly id: string field, collapsing resolveVoiceTtsLabel to a single ttsService.id property access and removing the per-provider localVoiceTtsProvider variable.
• ElevenLabsTTS uses constructor-injected fetch (via ElevenLabsTtsConfig.fetchImpl) for testability; the test suite covers URL building, path encoding, empty-input rejection, and error response mapping without vi.hoisted or global patching.
• persisted-config.ts widens the voice field from the OpenAI-specific enum to a plain string and extends SpeechProviderIdSchema with "elevenlabs", both documented as back-compat in the PR.

Confidence Score: 5/5

Safe to merge; the three new files form a contained module and the integration touches only the expected wiring points.

The feature core is cleanly isolated, the initialization pipeline is unchanged in structure, all previous review concerns have been addressed, and the remaining findings are minor naming and defensive-code nits that do not affect correctness.

tts.ts has two minor issues worth a second look: the misleading DEFAULT_ELEVENLABS_VOICE_ID export and the fetchImpl test seam surfacing in the public PaseoElevenLabsConfig type.

Important Files Changed

Filename	Overview
packages/server/src/server/speech/providers/elevenlabs/tts.ts	Core ElevenLabsTTS class; correctly injects fetch, uses Readable.fromWeb for streaming, handles error parsing. Minor: DEFAULT_ELEVENLABS_VOICE_ID exported but never used as a default; fetchImpl test seam exposed in public config type.
packages/server/src/server/speech/providers/elevenlabs/runtime.ts	Availability check, credential validation, and service initialization for ElevenLabs. Minor: redundant && elevenlabsConfig guard when hasTts already proves it non-null.
packages/server/src/server/speech/providers/elevenlabs/config.ts	Config resolver wires persisted-config and env vars into ElevenLabsTtsConfig; priority order (persisted > env) is correct and consistent with other providers.
packages/server/src/server/speech/speech-runtime.ts	ElevenLabs slot wired into the initialization pipeline (local → openai → elevenlabs); resolveVoiceTtsLabel simplified to use ttsService.id; localVoiceTtsProvider variable removed cleanly.
packages/server/src/server/speech/speech-provider.ts	TextToSpeechProvider gains required readonly id: string; all three TTS implementations (OpenAI, local, ElevenLabs) now set it, collapsing the label function to a single property access.
packages/server/src/server/speech/providers/elevenlabs/tts.test.ts	Tests use fetchImpl injection (no vi.hoisted / vi.stubGlobal); covers URL building, path encoding, empty-input rejection, and error response mapping.
packages/server/src/server/speech/providers/elevenlabs/runtime.test.ts	Tests cover availability predicate, service construction when provider matches, passthrough when it doesn't, and no-throw for valid configuration.
packages/server/src/server/persisted-config.ts	Adds ElevenLabsProviderSchema, extends SpeechProviderIdSchema with "elevenlabs", and widens voice field from OpenAI enum to string — documented as back-compat in the PR.

Sequence Diagram

%%{init: {'theme': 'neutral'}}%%
sequenceDiagram
    participant Config as resolveElevenLabsSpeechConfig
    participant Runtime as createSpeechService
    participant ELRuntime as initializeElevenLabsSpeechServices
    participant TTS as ElevenLabsTTS
    participant API as ElevenLabs API

    Runtime->>Config: "{ env, persisted }"
    Config-->>Runtime: "ElevenLabsTtsConfig | undefined"

    Runtime->>ELRuntime: "{ providers, elevenlabsConfig, existing }"
    ELRuntime->>TTS: new ElevenLabsTTS(config, logger)
    ELRuntime-->>Runtime: ExistingSpeechServices (ttsService set)

    Note over Runtime: resolveVoiceTtsLabel → ttsService.id = "elevenlabs"

    Runtime->>TTS: synthesizeSpeech(text)
    TTS->>API: "POST /v1/text-to-speech/{voiceId}?output_format=mp3_44100_128"
    API-->>TTS: "ReadableStream<Uint8Array> (MP3)"
    TTS-->>Runtime: "{ stream: Readable, format: "mp3" }"

Loading

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
sequenceDiagram
    participant Config as resolveElevenLabsSpeechConfig
    participant Runtime as createSpeechService
    participant ELRuntime as initializeElevenLabsSpeechServices
    participant TTS as ElevenLabsTTS
    participant API as ElevenLabs API

    Runtime->>Config: "{ env, persisted }"
    Config-->>Runtime: "ElevenLabsTtsConfig | undefined"

    Runtime->>ELRuntime: "{ providers, elevenlabsConfig, existing }"
    ELRuntime->>TTS: new ElevenLabsTTS(config, logger)
    ELRuntime-->>Runtime: ExistingSpeechServices (ttsService set)

    Note over Runtime: resolveVoiceTtsLabel → ttsService.id = "elevenlabs"

    Runtime->>TTS: synthesizeSpeech(text)
    TTS->>API: "POST /v1/text-to-speech/{voiceId}?output_format=mp3_44100_128"
    API-->>TTS: "ReadableStream<Uint8Array> (MP3)"
    TTS-->>Runtime: "{ stream: Readable, format: "mp3" }"

Loading

_{Reviews (5): Last reviewed commit: "refactor(speech): drop redundant ElevenL..." | Re-trigger Greptile}

[greptile-apps]

Want your agent to iterate on Greptile's feedback? Try greploops.