diff --git a/include/engine/models/kokoro_tts/frontend.h b/include/engine/models/kokoro_tts/frontend.h index 7b66789bb..8e7e6f673 100644 --- a/include/engine/models/kokoro_tts/frontend.h +++ b/include/engine/models/kokoro_tts/frontend.h @@ -6,6 +6,7 @@ #include #include #include +#include #include namespace engine::models::kokoro_tts { @@ -31,14 +32,25 @@ KokoroFrontendSessionState resolve_kokoro_frontend_session_state( const std::optional & voice, const KokoroAssets & assets); +// `phoneme_override`, when present, is synthesized as-is INSTEAD of running the built-in +// G2P over `text`. It lets a caller with its own grapheme-to-phoneme stage — a lexicon the +// engine does not carry, a language it does not cover, a pronunciation the application has +// already shown its user — drive the model directly. `text` is still required and its +// language must still agree with the voice; only the phonemization is replaced. +// +// ⚠ OPTIONAL, NOT "EMPTY MEANS NO". An empty string is a caller that asked for an override +// and gave nothing, and is rejected; it must not quietly fall back to speaking `text`, which +// is what an empty-string sentinel did here before. KokoroSynthesisInput build_kokoro_synthesis_input( const runtime::Transcript & text, const KokoroFrontendSessionState & state, - const KokoroAssets & assets); + const KokoroAssets & assets, + std::optional phoneme_override = std::nullopt); int64_t estimate_kokoro_request_tokens( const runtime::SessionPreparationRequest & request, const KokoroFrontendSessionState & state, - const KokoroAssets & assets); + const KokoroAssets & assets, + std::optional phoneme_override = std::nullopt); } // namespace engine::models::kokoro_tts diff --git a/model_specs/kokoro_tts.json b/model_specs/kokoro_tts.json index a4b4faff3..f5b49c1bb 100644 --- a/model_specs/kokoro_tts.json +++ b/model_specs/kokoro_tts.json @@ -29,6 +29,12 @@ "required": false, "min": 0 }, + { + "name": "phonemes", + "type": "string_list", + "description": "Kokoro-vocabulary phonemes to synthesize directly, bypassing the built-in eSpeak-ng G2P. One entry per chunk: entries are rendered in order and merged into one result, so the caller chooses the split points that only its own G2P knows. Each entry must be at most 510 symbols and must not be empty -- an entry is a chunk to speak, so drop it rather than leaving it blank. Text is still required and its language must still match the voice.", + "required": false + }, { "name": "text_chunk_size", "type": "int", diff --git a/src/framework/model_spec/options.cpp b/src/framework/model_spec/options.cpp index cf0324d47..7fb518837 100644 --- a/src/framework/model_spec/options.cpp +++ b/src/framework/model_spec/options.cpp @@ -50,6 +50,7 @@ const std::unordered_map> & shared_ {"min_p", {"float"}}, {"matmul_weight_type", {"enum"}}, {"negative_prompt", {"string"}}, + {"phonemes", {"string_list"}}, {"num_beams", {"int"}}, {"num_inference_steps", {"int"}}, {"output_sample_rate", {"int"}}, diff --git a/src/models/kokoro_tts/frontend.cpp b/src/models/kokoro_tts/frontend.cpp index 6bdfa82ac..4174c483d 100644 --- a/src/models/kokoro_tts/frontend.cpp +++ b/src/models/kokoro_tts/frontend.cpp @@ -117,9 +117,22 @@ struct EncodedInputIds { size_t phoneme_count = 0; }; +/// `reject_unknown` decides what an out-of-vocabulary symbol means, and the answer depends +/// entirely on who produced the stream. +/// +/// ⚠ FALSE for our own G2P, TRUE for a caller's. The reference implementation (hexgrad/Kokoro's +/// KModel, `filter(None, map(vocab.get, phonemes))`) drops what it cannot tokenize, and for our +/// own output that is the only sane answer: eSpeak emits a syllabic mark for "button" that this +/// vocabulary has no id for, and nobody downstream can do anything about it. +/// +/// A CALLER'S stream is the opposite case. They can fix it, so telling them is strictly more +/// useful than guessing -- and guessing is not harmless here. Canonical IPA writes a diphthong +/// as two symbols, and Kokoro writes it as one, so dropping the off-glide silently turns +/// `lˈaᶦk` into `lˈak`: "like" becomes "lack", with no error and audio that sounds deliberate. EncodedInputIds encode_input_ids_and_count( const std::string & phonemes, - const KokoroAssets & assets) { + const KokoroAssets & assets, + bool reject_unknown) { EncodedInputIds encoded; encoded.ids.reserve(phonemes.size() + 2); encoded.ids.push_back(0); @@ -149,6 +162,13 @@ EncodedInputIds encode_input_ids_and_count( const std::string symbol = phonemes.substr(i, width); const auto it = assets.vocab.find(symbol); if (it == assets.vocab.end()) { + if (reject_unknown) { + throw std::runtime_error( + "Kokoro vocab is missing phoneme symbol: " + symbol + + "; supplied phonemes must be in Kokoro's own " + std::to_string(assets.vocab.size()) + + "-symbol vocabulary, which is not canonical IPA -- a diphthong is one symbol there" + " and two in IPA, so an off-glide is a common cause"); + } // Skipped, not fatal, matching the reference implementation: hexgrad/Kokoro's KModel // tokenizes with `filter(None, map(vocab.get, phonemes))`, which drops any phoneme the // 114-entry vocab has no id for. @@ -231,15 +251,33 @@ KokoroFrontendSessionState resolve_kokoro_frontend_session_state( KokoroSynthesisInput build_kokoro_synthesis_input( const runtime::Transcript & text, const KokoroFrontendSessionState & state, - const KokoroAssets & assets) { + const KokoroAssets & assets, + std::optional phoneme_override) { if (state.voice_pack == nullptr) { throw std::runtime_error("Kokoro frontend session voice pack was not prepared"); } - const std::string phonemes = phonemize_text(text, state.language_code, assets); - const EncodedInputIds encoded = encode_input_ids_and_count(phonemes, assets); + const bool supplied = phoneme_override.has_value(); + if (supplied && phoneme_override->empty()) { + // Falling back to the built-in G2P here would speak `text` for a chunk the caller + // asked to be spoken from phonemes, which reads as the engine ignoring the option. + throw std::runtime_error("Kokoro supplied phonemes are empty; an override must carry symbols"); + } + const std::string phonemes = + supplied ? std::string(*phoneme_override) : phonemize_text(text, state.language_code, assets); + // Strict for a caller's stream, lenient for our own -- see encode_input_ids_and_count. + const EncodedInputIds encoded = encode_input_ids_and_count(phonemes, assets, supplied); if (encoded.phoneme_count > 510) { + // Two different failures wearing one message helps nobody: the caller who supplied the + // phonemes can fix this by sending less, and is told so; the caller who supplied text + // is hitting an engine limitation and is told that instead. throw std::runtime_error( - "Kokoro phoneme string exceeds 510 symbols; segmenting is not implemented in the framework path yet"); + supplied + ? "Kokoro phoneme chunk exceeds 510 symbols; split the supplied phonemes across " + "more list entries (each entry is rendered separately and the audio is merged)" + : "Kokoro phoneme string exceeds 510 symbols; segmenting is not implemented in the framework path yet"); + } + if (supplied) { + engine::debug::trace_log_scalar("kokoro.supplied_phoneme_count", static_cast(encoded.phoneme_count)); } KokoroSynthesisInput input; input.voice_id = state.voice_id; @@ -257,11 +295,12 @@ KokoroSynthesisInput build_kokoro_synthesis_input( int64_t estimate_kokoro_request_tokens( const runtime::SessionPreparationRequest & request, const KokoroFrontendSessionState & state, - const KokoroAssets & assets) { + const KokoroAssets & assets, + std::optional phoneme_override) { if (!request.text.has_value()) { return 0; } - const auto input = build_kokoro_synthesis_input(*request.text, state, assets); + const auto input = build_kokoro_synthesis_input(*request.text, state, assets, phoneme_override); return static_cast(input.input_ids.size()); } diff --git a/src/models/kokoro_tts/session.cpp b/src/models/kokoro_tts/session.cpp index 0547d50c7..a21577a86 100644 --- a/src/models/kokoro_tts/session.cpp +++ b/src/models/kokoro_tts/session.cpp @@ -13,7 +13,11 @@ #include #include #include +#include #include +#include +#include +#include namespace engine::models::kokoro_tts { @@ -236,8 +240,107 @@ void KokoroTTSSession::prepare_decoder_graph_capacity(int64_t capacity) { engine::debug::timing_log_scalar("kokoro.prepare.decoder_runtime_build_ms", build_ms); } +namespace { + +constexpr const char * kPhonemesOption = "phonemes"; + +/// The supplied phoneme chunks, or empty when the caller set no such option. +/// +/// ⚠ SET-BUT-EMPTY IS NOT THE SAME AS UNSET, at either level. An empty list, and an empty entry +/// within a list, both used to read as "no override" -- the first because the vector is empty, +/// the second because the frontend took an empty string as its sentinel -- and the chunk was +/// then spoken from `text`. A caller that asked for phonemes and supplied none got the source +/// text read aloud, which is the one thing this option exists to avoid. Both are refused here, +/// where the offending entry can still be named. The C ABI accepts a zero-length array +/// (audiocpp_request_set_option_array with count == 0), so the empty list is reachable. +const std::vector & require_phoneme_chunks( + const std::unordered_map> & option_arrays) { + static const std::vector none; + const auto it = option_arrays.find(kPhonemesOption); + if (it == option_arrays.end()) return none; + if (it->second.empty()) { + throw std::runtime_error( + "Kokoro 'phonemes' was set but holds no entries; pass the phonemes to synthesize, or " + "leave the option unset to use the built-in G2P"); + } + for (size_t index = 0; index < it->second.size(); ++index) { + if (it->second[index].empty()) { + throw std::runtime_error( + "Kokoro supplied phoneme entry " + std::to_string(index) + + " is empty; every entry is rendered as its own chunk, so each one must carry " + "symbols -- drop the entry instead of leaving it blank"); + } + } + return it->second; +} + +/// Builds one chunk's input, naming the entry a supplied-phoneme failure came from. +/// +/// The frontend's own rejections -- an out-of-vocab symbol, a chunk over 510 symbols -- say what +/// is wrong but not which entry it is in, and a list can be long. The empty-entry check above +/// names its entry, so these should too. +KokoroSynthesisInput build_supplied_or_text_input( + const runtime::TaskRequest & chunk_request, + const KokoroFrontendSessionState & state, + const KokoroAssets & assets, + std::optional chunk_phonemes, + size_t chunk_index, + bool supplied) { + if (!supplied) { + return build_kokoro_synthesis_input(*chunk_request.text_input, state, assets, chunk_phonemes); + } + try { + return build_kokoro_synthesis_input(*chunk_request.text_input, state, assets, chunk_phonemes); + } catch (const std::exception & error) { + throw std::runtime_error( + "Kokoro supplied phoneme entry " + std::to_string(chunk_index) + ": " + error.what()); + } +} + +/// The text-derived half of the synthesis cache key. +/// +/// Held apart from the phoneme half because in the supplied path every chunk shares one request: +/// this is the same string for all of them, and it carries the WHOLE document, so building it +/// per chunk would copy the caller's input once per entry. +std::string cache_key_prefix( + const KokoroFrontendSessionState & state, + const runtime::Transcript & text) { + return state.voice_id + ":" + + state.language_code + ":" + + std::to_string(state.speaking_rate) + ":" + + std::to_string(text.text.size()) + ":" + + text.text + ":"; +} + +/// Request options, validated against the package's own contract. +/// +/// Older standalone GGUF packages embed a schema-v1 contract written before +/// `phonemes` existed, and a published package cannot be edited in place. Drop +/// the key from the VALIDATION COPY so those packages can still be given +/// phonemes -- the engine serves the request either way -- while every unrelated +/// unknown option is still rejected. Same shape as irodori_tts.codec_backend and +/// the Parakeet TDT VAD controls, which are older options in the same position. +void validate_request_options( + const std::unordered_map & options, + const std::unordered_map> & option_arrays, + const engine::model_spec::ModelContract & contract) { + if (contract.request_option_keys.find(kPhonemesOption) != contract.request_option_keys.end()) { + runtime::validate_spec_backed_request_options(options, option_arrays, contract, kModelName); + return; + } + // Keys only: the validator reads names, and copying the map wholesale would copy every + // supplied phoneme string to drop one entry from it. + std::unordered_map> validation_arrays; + for (const auto & [key, _] : option_arrays) { + if (key != kPhonemesOption) validation_arrays.emplace(key, std::vector{}); + } + runtime::validate_spec_backed_request_options(options, validation_arrays, contract, kModelName); +} + +} // namespace + void KokoroTTSSession::prepare(const runtime::SessionPreparationRequest & request) { - runtime::validate_spec_backed_request_options(request.options, *contract_, kModelName); + validate_request_options(request.options, request.option_arrays, *contract_); if (const auto seed = runtime::parse_u64_option(request.options, {"seed"})) { if (rng_seed_ != *seed) { rng_seed_ = *seed; @@ -248,18 +351,40 @@ void KokoroTTSSession::prepare(const runtime::SessionPreparationRequest & reques } auto adapter = make_graph_capacity_adapter(); int64_t request_size = 0; + const auto & prepare_phonemes = require_phoneme_chunks(request.option_arrays); if (request.text.has_value()) { const int64_t text_chunk_size = engine::text::parse_text_chunk_size_override(request.options).value_or(kDefaultTextChunkSize); - const auto text_chunks = engine::text::split_text_chunks(request.text->text, text_chunk_size); + // The graph is sized for the LARGEST chunk either way. With supplied phonemes the + // caller's own chunking decides that, so the text is not split -- splitting it would + // size the graph against a boundary run() is never going to use. + const auto text_chunks = prepare_phonemes.empty() + ? engine::text::split_text_chunks(request.text->text, text_chunk_size) + : std::vector{request.text->text}; for (const auto & chunk : text_chunks) { runtime::SessionPreparationRequest chunk_request = request; chunk_request.text = runtime::Transcript{chunk, request.text->language}; const auto frontend_state = resolve_kokoro_frontend_session_state(chunk_request.text, chunk_request.voice, *assets_); - request_size = std::max( - request_size, - estimate_kokoro_request_tokens(chunk_request, frontend_state, *assets_)); + if (prepare_phonemes.empty()) { + request_size = std::max( + request_size, + estimate_kokoro_request_tokens(chunk_request, frontend_state, *assets_)); + } else { + // Every entry is sized here, before anything is synthesized, so a bad entry + // near the end of a long list is reported without rendering the ones before it. + for (size_t index = 0; index < prepare_phonemes.size(); ++index) { + try { + request_size = std::max( + request_size, + estimate_kokoro_request_tokens( + chunk_request, frontend_state, *assets_, prepare_phonemes[index])); + } catch (const std::exception & error) { + throw std::runtime_error( + "Kokoro supplied phoneme entry " + std::to_string(index) + ": " + error.what()); + } + } + } } } graph_capacity_controller_.ensure_prepared(adapter, request_size); @@ -271,34 +396,68 @@ runtime::TaskResult KokoroTTSSession::run(const runtime::TaskRequest & request) if (!request.text_input.has_value()) { throw std::runtime_error("Kokoro TTS run requires text_input"); } - runtime::validate_spec_backed_request_options(request.options, *contract_, kModelName); + validate_request_options(request.options, request.option_arrays, *contract_); const int64_t text_chunk_size = engine::text::parse_text_chunk_size_override(request.options).value_or(kDefaultTextChunkSize); - const auto chunk_requests = runtime::chunk_text_request(request, text_chunk_size); + // ⚠ WHEN PHONEMES ARE SUPPLIED THE CALLER OWNS THE CHUNKING. Chunking splits the TEXT, and + // nothing here knows where the matching cut points in someone else's phoneme stream are -- + // only their G2P does. So the option is a LIST: one entry per chunk, rendered in order and + // merged into one result exactly as text chunks are. A caller whose document exceeds the + // 510-symbol limit therefore still makes ONE call and gets ONE buffer back, instead of + // having to stitch the audio itself. + const auto & supplied_phonemes = require_phoneme_chunks(request.option_arrays); + // ⚠ NOT one TaskRequest per chunk. With supplied phonemes every chunk runs the SAME request + // -- only the phoneme entry differs -- and the request carries the whole phoneme list, so a + // copy per chunk would copy the caller's own input once for every entry in it. + const auto text_chunk_requests = supplied_phonemes.empty() + ? runtime::chunk_text_request(request, text_chunk_size) + : std::vector{}; + const size_t chunk_count = + supplied_phonemes.empty() ? text_chunk_requests.size() : supplied_phonemes.size(); engine::debug::trace_log_scalar("kokoro.text_chunk_size", text_chunk_size); - engine::debug::trace_log_scalar("kokoro.text_chunk_count", static_cast(chunk_requests.size())); + engine::debug::trace_log_scalar("kokoro.text_chunk_count", static_cast(chunk_count)); + if (!supplied_phonemes.empty()) { + engine::debug::trace_log_scalar( + "kokoro.supplied_phoneme_chunks", static_cast(supplied_phonemes.size())); + } double frontend_ms = 0.0; double inference_ms = 0.0; double predictor_ms = 0.0; double decoder_ms = 0.0; runtime::AudioBuffer merged_audio; - for (const auto & chunk_request : chunk_requests) { - const auto frontend_state = - resolve_kokoro_frontend_session_state(chunk_request.text_input, chunk_request.voice, *assets_); + // Resolved once in the supplied path, where every chunk runs the same request: the state + // and the text half of the key cannot differ between chunks there. + const bool supplied = !supplied_phonemes.empty(); + std::optional shared_state; + std::string shared_key_prefix; + if (supplied) { + shared_state = resolve_kokoro_frontend_session_state(request.text_input, request.voice, *assets_); + shared_key_prefix = cache_key_prefix(*shared_state, *request.text_input); + } + for (size_t chunk_index = 0; chunk_index < chunk_count; ++chunk_index) { + const auto & chunk_request = supplied ? request : text_chunk_requests[chunk_index]; + const std::optional chunk_phonemes = + supplied ? std::optional{supplied_phonemes[chunk_index]} + : std::optional{}; + const auto frontend_state = supplied + ? *shared_state + : resolve_kokoro_frontend_session_state(chunk_request.text_input, chunk_request.voice, *assets_); const std::string cache_key = - frontend_state.voice_id + ":" + - frontend_state.language_code + ":" + - std::to_string(frontend_state.speaking_rate) + ":" + - std::to_string(chunk_request.text_input->text.size()) + ":" + - chunk_request.text_input->text; + (supplied ? shared_key_prefix : cache_key_prefix(frontend_state, *chunk_request.text_input)) + + // Without this, two requests with the same text and different supplied phonemes + // would hit the same cache entry and the second would be spoken as the first -- + // and with a list, every chunk shares one text, so this is the ONLY thing telling + // them apart. + std::string(chunk_phonemes.value_or(std::string_view{})); KokoroSynthesisInput input; frontend_ms += measure_ms([&]() { if (!cache_key.empty() && cached_input_ && cache_key == cached_request_key_) { input = *cached_input_; return; } - input = build_kokoro_synthesis_input(*chunk_request.text_input, frontend_state, *assets_); + input = build_supplied_or_text_input( + chunk_request, frontend_state, *assets_, chunk_phonemes, chunk_index, supplied); if (!cache_key.empty()) { cached_request_key_ = cache_key; cached_input_ = std::make_unique(input);