{"id":18506,"date":"2026-11-18T01:10:25","date_gmt":"2026-11-17T16:10:25","guid":{"rendered":"https:\/\/localmodelwatch.tsuchitsuchi.com\/2026\/11\/18\/fishaudio-s2-pro-gguf-release\/"},"modified":"2026-11-18T01:10:25","modified_gmt":"2026-11-17T16:10:25","slug":"fishaudio-s2-pro-gguf-release","status":"publish","type":"post","link":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/18\/fishaudio-s2-pro-gguf-release\/","title":{"rendered":"s2-pro-GGUF 80-Language Text-to-Speech Model: 8GB+ VRAM, File List"},"content":{"rendered":"<p><!-- lmw:facts --><\/p>\n<h2>At a Glance<\/h2>\n<div class=\"lmw-table-scroll\" tabindex=\"0\" style=\"overflow-x:auto;-webkit-overflow-scrolling:touch;max-width:100%;\">\n<table style=\"width:max-content;min-width:100%;border-collapse:collapse;\">\n<thead>\n<tr>\n<th>Item<\/th>\n<th>Value<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Repository<\/td>\n<td><a href=\"https:\/\/huggingface.co\/ggml-org\/s2-pro-GGUF\">ggml-org\/s2-pro-GGUF<\/a><\/td>\n<\/tr>\n<tr>\n<td>Publisher guide<\/td>\n<td><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/publisher-fishaudio-en\/\">Fish Audio: models and licenses<\/a><\/td>\n<\/tr>\n<tr>\n<td>Published<\/td>\n<td>2026-11-18<\/td>\n<\/tr>\n<tr>\n<td>License<\/td>\n<td>fish-audio-research-license<\/td>\n<\/tr>\n<tr>\n<td>Formats<\/td>\n<td><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/format-gguf-en\/\">GGUF<\/a><\/td>\n<\/tr>\n<tr>\n<td>Source type<\/td>\n<td>Primary source (the publisher itself)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p><em>Values determined by this site&#8217;s code when the information was collected. Dates are JST.<\/em><\/p>\n<p><!-- \/lmw:facts --><\/p>\n<h2>Overview<\/h2>\n<p><code>ggml-org\/s2-pro-GGUF<\/code> is a GGUF conversion of the <code>fishaudio\/s2-pro<\/code> text-to-speech model. It is automatically converted and published by ggml-org and can be run using <code>llama-tts<\/code> or the <code>\/tts<\/code> endpoint in <code>llama serve<\/code>. The performance and specifications below are based on the original <code>fishaudio\/s2-pro<\/code> model card unless otherwise noted.<\/p>\n<h2>Specifications<\/h2>\n<ul>\n<li>Original Model: <code>fishaudio\/s2-pro<\/code> (Fish Audio S2 Pro). The model card states it was trained on &#8220;10M+ hours of audio data across 80+ languages.&#8221;<\/li>\n<li>Architecture: A Dual-Autoregressive (Dual-AR) configuration combining a decoder-only Transformer with an RVQ-based audio codec (10 codebooks, approx. 21Hz frame rate).<\/li>\n<li>Slow AR (4B parameters): Operates on the time axis to predict the primary semantic codebook. &#8211; Fast AR (400M parameters): Generates the remaining 9 residual codebooks at each time step to reconstruct fine acoustic details.<\/li>\n<li>Training Method: Reported to incorporate alignment via reinforcement learning.<\/li>\n<li>Supported Languages: 80+<\/li>\n<li>Tier 1: Japanese (ja), English (en), Chinese (zh) &#8211; Tier 2: Korean, Spanish, Portuguese, Arabic, Russian, French, German &#8211; Also supports many others including sv, it, tr, no, nl, etc.<\/li>\n<li>Inline Control: Natural language instructions can be embedded in the text using <code>[tag]<\/code> format. It accepts free-form descriptions like <code>[whisper in small voice]<\/code>, <code>[professional broadcast tone]<\/code>, or <code>[pitch up]<\/code> rather than a fixed set of tags. Frequently used tags include <code>[pause]<\/code>, <code>[emphasis]<\/code>, <code>[laughing]<\/code>, <code>[whisper]<\/code>, <code>[sad]<\/code>, and <code>[singing]<\/code>, with support for over 15,000 unique tags.<\/li>\n<li>GGUF Usage Features:<\/li>\n<li>Assign voices to characters using speaker tags (e.g., <code>&lt;|speaker:0|&gt;<\/code>). &#8211; For voice cloning, specify a reference audio file via <code>--tts-speaker-file<\/code> and provide a transcription via <code>--tts-speaker-text<\/code> if necessary.<\/li>\n<li>License Terms: According to the GGUF model card, it is free for research and non-commercial use, with restrictions on commercial use. Please check the full license text for details.<\/li>\n<\/ul>\n<h2>Performance<\/h2>\n<p>There are no comparison tables or benchmark scores against other models in the model card. The following information is provided by the original developer (Fish Audio) regarding the base model:<\/p>\n<ul>\n<li>The model card positions S2 Pro as an &#8220;advanced TTS model capable of fine-grained control over prosody and emotion at the word level.&#8221;<\/li>\n<li>Streaming performance on a single NVIDIA H200 is reported as follows:<\/li>\n<li>Real-Time Factor (RTF): 0.195 &#8211; Time to first audio: approx. 100 ms &#8211; Throughput: 3,000+ acoustic tokens\/s while keeping RTF below 0.5<\/li>\n<\/ul>\n<p>These figures are based on an SGLang-based streaming inference engine. Speed and quality when running the GGUF version via <code>llama-tts<\/code> are not documented and remain unverified.<\/p>\n<h2>Strengths and Use Cases<\/h2>\n<p>Based on the <code>fishaudio\/s2-pro<\/code> model card:<\/p>\n<ul>\n<li><strong>Fine-grained emotional and prosodic control<\/strong>: Tags like <code>[laughing]<\/code>, <code>[whisper]<\/code>, <code>[sad]<\/code>, <code>[excited]<\/code>, and <code>[pause]<\/code> can be inserted at the word level. Free-form descriptions like <code>[whisper in small voice]<\/code> or <code>[professional broadcast tone]<\/code> are also supported. This is ideal for narration, character dialogue, and conversational audio requiring expressive intonation.<\/li>\n<li><strong>Multi-speaker interaction<\/strong>: The GGUF version allows assigning voices to different characters using speaker tags like <code>&lt;|speaker:0|&gt;<\/code>.<\/li>\n<li><strong>Voice cloning<\/strong>: Use <code>--tts-speaker-file<\/code> to provide a reference audio. If a transcription is available, it can be provided via <code>--tts-speaker-text<\/code>. The fidelity of the voice cloning is not documented and remains unverified.<\/li>\n<li><strong>Multilingual synthesis<\/strong>: Supports over 80 languages, with Japanese, English, and Chinese in Tier 1. This is good news for users wanting to test Japanese speech synthesis locally.<\/li>\n<\/ul>\n<p>This GGUF version is an automatic conversion of the original model weights by ggml-org. It is intended to enable execution via <code>llama.cpp<\/code> tools rather than as a fine-tuned model with altered performance or capabilities.<\/p>\n<p>If you are considering commercial use, please be cautious. The GGUF card states that it is free for research and non-commercial use, but commercial use is restricted. Please verify the license terms in detail.<\/p>\n<p><!-- lmw:hardware --><\/p>\n<h2>Hardware Requirements<\/h2>\n<p><strong>Estimated requirements (calculated by Local Model Watch)<\/strong> \u2014 4.6B parameters (taken from the base model fishaudio\/s2-pro)<\/p>\n<div class=\"lmw-table-scroll\" tabindex=\"0\" style=\"overflow-x:auto;-webkit-overflow-scrolling:touch;max-width:100%;\">\n<table style=\"width:max-content;min-width:100%;border-collapse:collapse;\">\n<thead>\n<tr>\n<th>Your VRAM<\/th>\n<th>Quantization<\/th>\n<th>File size<\/th>\n<th>Est. memory needed<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>8GB (RTX 4060 \/ 3060 Ti, etc.)<\/td>\n<td>Q8_0<\/td>\n<td>4.0GB<\/td>\n<td>4.8GB<\/td>\n<\/tr>\n<tr>\n<td>12GB (RTX 4070 \/ 3060 12GB, etc.)<\/td>\n<td>BF16<\/td>\n<td>7.5GB<\/td>\n<td>9.0GB<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p><em>Memory estimates add a 20% runtime overhead (KV cache, etc.) to the actual size of the distributed files. Actual usage varies with context length, batch size and inference engine. These figures are computed by this site from file sizes, not published by the model&#8217;s authors. Compare with other models in our <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/vram-guide-en\/\">VRAM quick reference<\/a>. What the quantization names mean: <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/glossary-quantization-en\/\">glossary<\/a>.<\/em><\/p>\n<p><!-- \/lmw:hardware --><\/p>\n<p><!-- lmw:runnability --><\/p>\n<h2>Can You Run It Locally?<\/h2>\n<p>The publisher distributes this model as GGUF.<\/p>\n<p><strong>License \u2014 <code>fish-audio-research-license<\/code>:<\/strong> A custom license from the publisher. Check the original terms directly, including whether commercial use is permitted.<\/p>\n<p><strong>Compression:<\/strong> the Q8_0 build measures 7.53 bits per weight \u2014 about 47% the size of the original 16-bit weights, calculated by this site from the actual file sizes.<\/p>\n<p><em>Compiled by this site&#8217;s code from the published formats and the license field. License summaries are not legal advice \u2014 check the publisher&#8217;s original terms before relying on them.<\/em><\/p>\n<p><!-- \/lmw:runnability --><\/p>\n<p><!-- lmw:lab --><\/p>\n<h2>Our Own Measurements<\/h2>\n<h3>Measurements We Did Not Take<\/h3>\n<p>We have not confirmed that this site meets the commercial-use terms of this model&#8217;s license (<code>fish-audio-research-license<\/code>). Because this site carries advertising, we did not run the model (no CPU run, answers, quantization comparison, conversion or generation).<\/p>\n<p><!-- \/lmw:lab --><\/p>\n<p><!-- lmw:files --><\/p>\n<h2>Distributed Files<\/h2>\n<p><em>Weight files published in <a href=\"https:\/\/huggingface.co\/ggml-org\/s2-pro-GGUF\/tree\/main\">ggml-org\/s2-pro-GGUF<\/a>, listed by this site from the Hugging Face API. Sizes are the actual file sizes.<\/em><\/p>\n<div class=\"lmw-table-scroll\" tabindex=\"0\" style=\"overflow-x:auto;-webkit-overflow-scrolling:touch;max-width:100%;\">\n<table style=\"width:max-content;min-width:100%;border-collapse:collapse;\">\n<thead>\n<tr>\n<th>File<\/th>\n<th>Size<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><code>mmproj-s2-pro-BF16.gguf<\/code><\/td>\n<td>2.05GB<\/td>\n<\/tr>\n<tr>\n<td><code>s2-pro-BF16.gguf<\/code><\/td>\n<td>8.07GB<\/td>\n<\/tr>\n<tr>\n<td><code>s2-pro-Q8_0.gguf<\/code><\/td>\n<td>4.29GB<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p><!-- \/lmw:files --><\/p>\n<h2>How to Get It<\/h2>\n<ul>\n<li>Distribution Format: GGUF. It is not gated, so you can download it from Hugging Face without needing to agree to a license on the site. However, the weights are released under the Fish Audio Research License.<\/li>\n<li>Repository: <code>ggml-org\/s2-pro-GGUF<\/code> (Original model: <code>fishaudio\/s2-pro<\/code>)<\/li>\n<li>Supported Tools: <code>llama-tts<\/code> and <code>llama serve<\/code> from <code>llama.cpp<\/code>. Support for other tools like ComfyUI is not mentioned in the documentation.<\/li>\n<\/ul>\n<p>To run from the terminal, use the following command:<\/p>\n<pre><code class=\"language-bash\">llama-tts -hf ggml-org\/s2-pro-GGUF -p &quot;&lt;|speaker:0|&gt;Hello world [laughing]&quot; --output out.wav\n<\/code><\/pre>\n<p>To use it as a server, start it with the following command and call the <code>\/tts<\/code> endpoint:<\/p>\n<pre><code class=\"language-bash\">llama serve -hf ggml-org\/s2-pro-GGUF\n<\/code><\/pre>\n<p>Specific instructions for voice cloning are summarized in the <code>llama.cpp<\/code> <a href=\"https:\/\/github.com\/ggml-org\/llama.cpp\/blob\/master\/docs\/multimodal\/tts.md#fish-audio-s2-pro\">TTS documentation (Fish Audio S2 Pro section)<\/a>.<\/p>\n<p>According to the card, this conversion is performed automatically by <a href=\"https:\/\/github.com\/ggml-org\/convert\">https:\/\/github.com\/ggml-org\/convert<\/a>. Differences in quality between quantization levels are not documented.<\/p>\n<p><!-- lmw:related --><\/p>\n<h2>Related Articles<\/h2>\n<ul>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/ggml-org-vibevoice-asr-gguf\/\">VibeVoice-ASR-GGUF Speech Recognition and Diarization Model: 8GB+ VRAM<\/a><\/li>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/minimax-music3-gguf-released\/\">MiniMax-Music3-GGUF Full-Length Music Generation Model: 12GB+ VRAM<\/a><\/li>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/higgs-audio-v3-stt-gguf\/\">higgs-audio-v3-stt-GGUF Speech-to-Text Model: 4GB+ VRAM, File List<\/a><\/li>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/soprano-1-1-80m-gguf-released\/\">Soprano-1.1-80M-GGUF Lightweight English TTS Model: File List<\/a><\/li>\n<\/ul>\n<p><!-- \/lmw:related --><\/p>\n<p><!-- lmw:next-steps --><\/p>\n<h2>What to Read Next<\/h2>\n<ul>\n<li><strong>Find models by VRAM<\/strong> (This model runs from the 8GB tier) \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/vram-guide-8gb-en\/\">Other models that run on a 8GB GPU<\/a><\/li>\n<li><strong>What Q8_0 mean and where to get this model<\/strong> \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/format-gguf-en\/\">GGUF format guide and models<\/a> \/ <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/glossary-quantization-en\/\">Quantization and model-format glossary<\/a><\/li>\n<li><strong>Learn about the publisher<\/strong> \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/publisher-fishaudio-en\/\">Fish Audio: models, licenses and articles<\/a><\/li>\n<li><strong>Other models for the same task<\/strong> \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/models-by-task-en\/#task-audio\">Other audio models<\/a><\/li>\n<\/ul>\n<p><!-- \/lmw:next-steps --><\/p>\n<h2>Sources<\/h2>\n<ul>\n<li><a href=\"https:\/\/huggingface.co\/ggml-org\/s2-pro-GGUF\">https:\/\/huggingface.co\/ggml-org\/s2-pro-GGUF<\/a><\/li>\n<li><a href=\"https:\/\/huggingface.co\/fishaudio\/s2-pro\">https:\/\/huggingface.co\/fishaudio\/s2-pro<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/ggml-org\/llama.cpp\/blob\/master\/docs\/multimodal\/tts.md#fish-audio-s2-pro\">https:\/\/github.com\/ggml-org\/llama.cpp\/blob\/master\/docs\/multimodal\/tts.md#fish-audio-s2-pro<\/a><\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>The text-to-speech model fishaudio\/s2-pro is now available in GGUF format, enabling local inference via llama.cpp tools like llama-tts and llama serve.<\/p>\n","protected":false},"author":1,"featured_media":18505,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[203],"tags":[5261,161,5263,163,1073,5265,1547,213],"class_list":["post-18506","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-image-video-and-audio","tag-fishaudio-s2-pro-en","tag-ggml-en","tag-ggml-org-s2-pro-gguf-en","tag-gguf-en","tag-llama-cpp-en","tag-s2-pro-en","tag-verified","tag--en"],"lang":"en","translations":{"en":18506,"ja":18504},"_links":{"self":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/18506","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/comments?post=18506"}],"version-history":[{"count":0,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/18506\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/media\/18505"}],"wp:attachment":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/media?parent=18506"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/categories?post=18506"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/tags?post=18506"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}