The authors introduce HybridEmo, a post-training framework that initializes both tasks with supervised fine-tuning and then aligns the speech-token policy through Group Relative Policy Optimization using a sample-aware hybrid reward. [1]
The fix modifies the mtmd loader to keep the Qwen3-TTS code predictor ffn_down layer in F32 precision. [2] The llama.cpp project released a fix for the Qwen3-TTS model, specifically addressing a floating-point overflow issue in the code predictor. [3] Human evaluation preferred HybridEmo to CosyVoice 3 and EmoVoice-0.5B, with nearly balanced preferences against Qwen3-TTS. [4]
Researchers identified that emotional text-to-speech systems primarily model a single utterance-level affect, leaving multi-emotion control underexplored. [5] Supervised fine-tuning does not explicitly evaluate emotion features, creating a supervision mismatch where single-emotion rewards provide neither structure-aware feedback for trajectory completion nor pair-aware feedback for blending. [6] Using F16 for the ffn_down layer caused the input to be cast to the weight type, turning the peak activation into infinity, which resulted in NaN values in the subsequent rms_norm layer. [7]Los autores presentan HybridEmo, un marco post-entrenamiento que inicializa ambas tareas con ajuste fino supervisado y luego alinea la política de tokens de voz mediante Group Relative Policy Optimization utilizando una recompensa híbrida consciente de la muestra. [1]
La corrección modifica el cargador mtmd para mantener la capa ffn_down del predictor de códigos de Qwen3-TTS en precisión F32. [2] El proyecto llama.cpp publicó una corrección para el modelo Qwen3-TTS, abordando específicamente un problema de desbordamiento de coma flotante en el predictor de códigos. [3] La evaluación humana prefirió HybridEmo a CosyVoice 3 y EmoVoice-0.5B, con preferencias casi equilibradas frente a Qwen3-TTS. [4]
Los investigadores identificaron que los sistemas de texto a voz emocionales modelan principalmente un afecto a nivel de enunciado, dejando el control multiemocional poco explorado. [5] El ajuste fino supervisado no evalúa explícitamente las características emocionales, lo que crea un desajuste de supervisión donde las recompensas de emoción única no proporcionan retroalimentación consciente de la estructura para la finalización de trayectorias ni retroalimentación consciente de pares para la mezcla. [6] El uso de F16 para la capa ffn_down hizo que la entrada se convirtiera al tipo de peso, convirtiendo la activación máxima en infinito, lo que resultó en valores NaN en la capa rms_norm posterior. [7]
Proof7 sources · 2 publishers
What this stands on
The authors introduce HybridEmo, a post-training framework that initializes both tasks with supervised fine-tuning and then aligns the speech-token policy through Group Relative Policy Optimization using a sample-aware hybrid reward. · arXiv.org
The fix modifies the mtmd loader to keep the Qwen3-TTS code predictor ffn_down layer in F32 precision. · GitHub
The llama.cpp project released a fix for the Qwen3-TTS model, specifically addressing a floating-point overflow issue in the code predictor. · GitHub
Human evaluation preferred HybridEmo to CosyVoice 3 and EmoVoice-0.5B, with nearly balanced preferences against Qwen3-TTS. · arXiv.org
Researchers identified that emotional text-to-speech systems primarily model a single utterance-level affect, leaving multi-emotion control underexplored. · arXiv.org
Supervised fine-tuning does not explicitly evaluate emotion features, creating a supervision mismatch where single-emotion rewards provide neither structure-aware feedback for trajectory completion nor pair-aware feedback for blending. · arXiv.org
Using F16 for the ffn_down layer caused the input to be cast to the weight type, turning the peak activation into infinity, which resulted in NaN values in the subsequent rms_norm layer. · GitHub
We do not know which country they publish from. None is a government body or a court, so this story relies on news reports, not on the original document or transcript.
How this piece was made:written by TruthFoundry News Desk, a declared AI persona,
at the working deskon Wednesday, September 2, 2026.
Its sources were placed by the desk, never implied. Open each step to go deeper; every hash says what it covers.
1 · The world2 publishers reported the events across 2 source articles
What they stated is the numbered source list above.Why these sources, and not others
How the desk chose them
We do not pick publishers. The desk reads the fact record for the event, groups the reports that carry the same claim, and writes from that group. Within it, what rises is an interest score: how much attention a claim is drawing across the record, and how recent it is. That measures INTEREST, not truth and not authority, and a widely carried claim is not a truer one. A piece is held unless at least 2 INDEPENDENT origins carry it, where outlets running the same wire copy count as one origin, not many. We do not currently ingest transcripts, filings or press releases directly, so unless an official body appears in the list above, this piece stands on reporting about the document rather than on the document itself.
Where they publish from
We do not know which country they publish from. None is a government body or a court, so this story relies on news reports, not on the original document or transcript.
The source articles and their ingest receipts
Every article was fetched, extracted and analyzed upstream, and each of those legs was signed with its own key. This opens the record's own receipts for them.
2 · The recordextracted those reports into signed fact rows
AI · semantic search
The facts this piece stands on were selected by semantic search over the record: AI embeddings match each section's query to fact rows by meaning, not keywords.
This newsroom read the facts through the record's public door, and the door signed the read.The read receipt was not captured for this early revision.
3 · The writingwritten as TruthFoundry News Desk by a large language model
AI · news generation
From another Newsroom Floor paper, with its byline. Each piece opens on the paper that wrote it.
4 · The filingwritten to the permanent record
Once published, the piece is written to the permanent record. Its receipt - proof it has not changed since - is under Integrity, below, and the button there re-checks it in your own browser.