What You'll Build
A local text-to-speech pipeline using hexgrad/Kokoro-82M — an 82-million-parameter Apache-2.0 TTS model that emits 24 kHz audio across 9 languages and 54 voices (counted from the canonical voices/ directory, 54 .pt tensor files, matching the model card's "8 & 54" header). Kokoro is wildly over-provisioned for a 12 GB RTX 5070: it needs only 2–3 GB total during inference, so the angle of this recipe is not "does it fit" (it fits roughly 6× over) — it is how to spend the spare ~10 GB by colocating Kokoro with a second LLM or generative workload on the same card.
Hardware data: RTX 5070 (12 GB VRAM) · weights fit under 1 GB at FP16; total inference footprint typically 2–3 GB · See benchmark data
Sizing note: Kokoro is borderline hardware-agnostic — it runs comfortably on a single RTX 3060 or even on CPU. The 12 GB RTX 5070 is wildly over-provisioned for this 82M model on its own (2–3 GB on 12 GB is roughly 6× over-provisioned). The install steps below apply unchanged to any modern NVIDIA GPU with >= 2 GB VRAM; the unique 5070 angle is the Colocating with a second model section. No FlashAttention-specific tweaks are required — Kokoro does not depend on FlashAttention-2, so the FA2 sm_120 kernel gap that bites larger Blackwell workloads is a non-issue here.
Requirements
| Component | Minimum | Tested |
|---|---|---|
| GPU | 2 GB VRAM (per Clore.ai guide) | RTX 5070 (12 GB) |
| RAM | 8 GB | — |
| Storage | ~1 GB (weights ~200 MB; rest is the Python wheel + misaki G2P data) | — |
| Software | Python 3.9+ (per Clore.ai), PyTorch with CUDA (cu128 build for Blackwell), espeak-ng | — |
Installation
1. Install espeak-ng at the OS level
The misaki G2P (grapheme-to-phoneme) library underneath Kokoro shells out to espeak-ng. Install it the system way for your OS — the Linux apt-get line is taken from the HF model card's Colab quickstart cell; the macOS Homebrew and Windows installer lines are the canonical equivalents:
# Linux (Debian / Ubuntu)
sudo apt-get install -y espeak-ng
# macOS
brew install espeak-ng
# Windows — download and run the .msi from
# https://github.com/espeak-ng/espeak-ng/releases
2. Install PyTorch with the Blackwell-compatible CUDA build
The RTX 5070 is a Blackwell (GB205, sm_120) card. Install the cu128 PyTorch wheel so the kernels match — cu126 and earlier wheels do not ship sm_120 kernels:
pip install torch --index-url https://download.pytorch.org/whl/cu128
Kokoro itself does not call FlashAttention-2, xformers, or sageattention, so the FA2 sm_120 kernel gap does not apply to this recipe — only the base PyTorch CUDA build needs to match the card.
3. Install the Python package
The kokoro PyPI package pulls in misaki, the model loader, and the inference loop. From the Hugging Face model card:
pip install "kokoro>=0.9.2" soundfile
Optional non-English language packs (only install what you need):
pip install "misaki[ja]" # Japanese
pip install "misaki[zh]" # Mandarin Chinese
4. (Optional) Pre-download the weights
The first call to KPipeline downloads ~200 MB of weights to your Hugging Face cache. Pre-fetch them if you want to control where they land:
from huggingface_hub import snapshot_download
snapshot_download("hexgrad/Kokoro-82M")
Running
The minimal Python example, verbatim from the Hugging Face model card:
from kokoro import KPipeline
import soundfile as sf
pipeline = KPipeline(lang_code='a') # 'a' = American English
text = "Kokoro is an open-weight 82-million-parameter text-to-speech model."
generator = pipeline(text, voice='af_heart')
for i, (gs, ps, audio) in enumerate(generator):
sf.write(f'{i}.wav', audio, 24000)
Note the 24000 sample rate — Kokoro emits 24 kHz audio. gs and ps are the graphemes and phonemes for the chunk, useful for debugging pronunciation.
Language codes
From the official GitHub README, lang_code is a single letter:
| Code | Language |
|---|---|
a | American English |
b | British English |
e | Spanish |
f | French |
h | Hindi |
i | Italian |
j | Japanese (requires misaki[ja]) |
p | Brazilian Portuguese |
z | Mandarin Chinese (requires misaki[zh]) |
The voice ID (af_heart above) selects one of the 54 voices — see the VOICES.md catalogue for the descriptions. The prefix encodes language + gender (e.g. af_ = American female).
Server-style deployment (optional)
If you'd rather expose an OpenAI-compatible HTTP API instead of writing Python, the community-maintained Kokoro-FastAPI wrapper is the most-cited path, per ThinkSmart.Life's local-rig writeup:
git clone https://github.com/remsky/Kokoro-FastAPI.git
cd Kokoro-FastAPI/docker/gpu
docker compose up --build
The repo also ships a top-level start-gpu.sh helper that wraps the same compose invocation.
Colocating with a second model (the real use case)
A 5070 running Kokoro alone leaves ~10 GB of unused VRAM. The legitimate per-GPU angle here is to keep Kokoro resident on one CUDA context and load a second model alongside it on the same 12 GB card. Concrete combinations that fit the envelope:
| Stack | Kokoro footprint | Companion | Companion footprint | Total resident |
|---|---|---|---|---|
| TTS + chat | 2–3 GB | Qwen3-8B Q4_K_M GGUF | ~6 GB peak | ~9 GB; ~3 GB free |
| TTS + vision chat | 2–3 GB | Gemma 4 E4B Q4 via llama.cpp | ~4.5 GB | ~7 GB; ~5 GB free |
Kokoro's own GPU footprint stays at the 2–3 GB Spheron reports for any of these combinations — it is the other model that defines whether the pair fits 12 GB. Run each companion in its own process / CUDA context (a separate python for the LLM via llama.cpp or Ollama, and the FastAPI wrapper above for Kokoro) so VRAM allocation stays additive and each component can be restarted independently. On a 12 GB card the companion choice is tighter than on a 16 GB card — a 20B image-gen model that fits a 16 GB sibling does not leave room for Kokoro here, so favour an 8B-class LLM or a vision model in the 4–6 GB band. For 16 GB-class headroom that fits a heavier companion alongside Kokoro, see the RTX 5070 Ti recipe.
Results
- Speed: No RTX 5070 speed numbers exist for Kokoro yet. Kokoro's tiny 82M architecture is faster-than-realtime on any modern consumer GPU; the Spheron deployment guide summarises a real-time factor (RTF) of about 0.03 measured on an A100 (i.e. 1 s of audio synthesised in ~30 ms). At this 82M scale the bottleneck is the G2P front-end and per-chunk Python overhead rather than raw tensor throughput, so the RTX 5070 will also be comfortably faster-than-realtime — but because no source has benchmarked Kokoro on the 5070 by name, track /check/kokoro-tts/rtx-5070 for empirical numbers and contribute your own via /contribute.
- VRAM usage: Weights are under 1 GB at FP16; total GPU memory during inference (including CUDA kernels and buffers) is 2–3 GB, per the Spheron deployment guide. The Clore.ai guide lists 2 GB as the minimum and 4 GB as the recommended VRAM, with an RTX 3060 as their recommended card — leaving the 12 GB RTX 5070 with ~10 GB of spare headroom for the colocation table above.
- Quality notes: 24 kHz output, 54 voices, 9 languages, Apache-2.0 license. Long text is chunked automatically by the pipeline iterator, which is why the
Runningexample loops overgenerator.
For the full benchmark data, see /check/kokoro-tts/rtx-5070.
Troubleshooting
RuntimeError: espeak-ng not found on first synthesis call
The Python kokoro package wraps misaki, which in turn calls espeak-ng for phonemisation. The PyPI install does not bundle the binary — you must install it through your OS package manager (apt-get install espeak-ng / brew install espeak-ng / Windows .msi) as covered in step 1. Per the official GitHub README.
PyTorch crash on first inference call (Blackwell-specific)
If you installed PyTorch via the default index (pip install torch), you may have landed a cu126 wheel that does not ship sm_120 kernels for the 5070. Reinstall with the explicit cu128 index per step 2:
pip uninstall -y torch
pip install torch --index-url https://download.pytorch.org/whl/cu128
Kokoro itself does not depend on FlashAttention-2, so the FA2 sm_120 kernel gap is not a concern here — only the base PyTorch CUDA build matters.
Non-English voice errors
Japanese (lang_code='j') and Mandarin (lang_code='z') require the optional misaki language packs — pip install "misaki[ja]" or pip install "misaki[zh]". Without them you'll see a missing-dependency traceback at pipeline init. Source: Hugging Face model card.
Pip install fails with misaki[en]>=X.Y.Z has no matching distribution
Reported in the upstream issue tracker — pin kokoro to a version whose declared misaki dependency is actually published on PyPI (the latest tagged release on the GitHub repo is the safe bet). Avoid installing from main if the bumped misaki requirement hasn't shipped yet.