EmoRES-TTS is the official implementation of the paper "EmoRES-TTS: Residual-Enhanced Vector Steering for Emotional Speech Generation" from Meta's facebookresearch organization. It is a steering method, not a standalone TTS model, and the repository contains no model weights. The method takes a mixed-emotion steering vector, decomposes it into a shared axis and an emotion-carrying residual, and weights the two parts independently when steering an existing TTS backbone. The code is built on top of CoCoEmo, an MIT-licensed project that is a required dependency and is not vendored here; the steering vectors, injection hooks, backbone adapters, and the TEP, rho, H-Rate, S-SIM, and WER metric implementations are imported from CoCoEmo rather than reimplemented. To run it you clone both repositories and supply a backbone checkpoint (IndexTTS-2 or CosyVoice2), CoCoEmo's steering vectors, emotion2vec_plus_large, and the ESD emotional speech dataset used for anchored E-SIM, with data/esd_anchors.txt listing the 400 anchor clips by path. The repository holds an emores Python package (compose, rule, esim, and a CosyVoice compatibility module) and scripts to build IEMOCAP and CREMA-D mixed-emotion manifests, synthesize with EmoRES, CoCoEmo-style raw steering, or no steering, and score anchored E-SIM. Evaluation uses IEMOCAP, with data/iemocap_1083_ids.txt listing the 1,083 utterance ids, and CREMA-D, which also needs its rater votes downloaded separately to define the mixed-emotion targets. No corpus audio, transcripts, or embeddings are redistributed, and each corpus must be obtained under its own terms. The anchored E-SIM formulation follows EmoSteer-TTS. Results are reported in the arXiv paper (2609.38157) and on the project page. The code is licensed CC BY-NC 4.0, which allows non-commercial use only, and third-party models and corpora keep their own licences.