Summary

Closes #10164.

The OpenAI-compatible TTS endpoint (POST /v1/audio/speech) accepts an instructions field, but it was silently dropped at the HTTP→gRPC boundary: neither schema.TTSRequest nor the gRPC TTSRequest proto carried it, so backends could only read such a value from static YAML options (identical for every request). This blocked:

• Per-line emotion/style (Qwen3-TTS CustomVoice, Chatterbox): same speaker, different tone per request.
• VoiceDesign / describe-a-voice (Qwen3-TTS VoiceDesign): the instruction string is the voice, so a model config was limited to a single designed voice.

This PR plumbs a generic per-request instruction string end to end, plus an optional backend-specific params map.

Changes

• proto (backend/backend.proto): add optional string instructions = 6 and map<string,string> params = 7 to TTSRequest.
• schema (core/schema/localai.go): add Instructions (maps the OpenAI instructions field) and Params (LocalAI extension).
• core (core/backend/tts.go): thread both through ModelTTS/ModelTTSStream via a newTTSRequest helper that attaches instructions only when non-empty (so backends can fall back to YAML when unset). Forwarded from the /v1/audio/speech handler; other callers (cli, elevenlabs, realtime) pass empty values.
• qwen-tts (backend/python/qwen-tts/backend.py): prefer the per-request instruction over the YAML instruct option (used by both mode detection and generation) and merge per-request params.
• chatterbox (backend/python/chatterbox/backend.py): merge per-request params (coerced to float/int/bool) over YAML options into generate() kwargs.
• docs + regenerated swagger.

Backward compatibility

Fully compatible: empty instructions falls back to the YAML option, and backends that don't support style/voice instructions simply ignore the field. The params map values arrive as strings and are coerced by the backend.

Example

curl http://localhost:8080/v1/audio/speech -H "Content-Type: application/json" -d '{
  "model": "qwen-tts-design",
  "input": "Hello world, this is a test.",
  "instructions": "A calm, low-pitched elderly storyteller with a warm tone."
}'

Testing

• New Ginkgo specs for newTTSRequest (instructions attached/omitted, params forwarded/nil); updated existing ctx-propagation specs for the new signature.
• core/backend, core/schema, core/http/endpoints/localai test suites pass.
• go vet and golangci-lint --new-from-merge-base=master clean (0 issues); both Python backends py_compile clean.

Generated proto bindings (pkg/grpc/proto/*) are gitignored and rebuilt by CI; Python backend_pb2 is regenerated from backend.proto at backend build time.

Assisted-by: Claude:claude-opus-4-8 [Claude Code]