--- title: "Miso TTS Self-Hosted vs API: Cost and Latency" canonical: https://wavect.io/blog/miso-tts-self-hosted-vs-api/ language: en description: "Compare Miso TTS self-hosting vs its API on 110 ms latency, 24 GB VRAM, pricing, license, privacy and voice-agent production readiness." image: "https://wavect.io/img/blog/headers/header_miso-tts-self-hosted-vs-api.png" --- [**Back**](/blog/overview/) [![Kevin Riedl](/img/team/kevin.webp)](/team/kevin-riedl/) [Kevin Riedl](/team/kevin-riedl/) https://linkedin.com/in/wsdt 11 min read · 26 Jul 2026 Last reviewed July 26, 2026 [**Next**](/blog/nvidia-nemotron-3-5-asr-production-review/) # Miso TTS Self-Hosted vs API: Cost, Latency and VRAM Reality TL;DR Miso One is the product-facing name people use for Miso Labs' Miso TTS 8B. The downloadable model can condition on short prompt audio for voice cloning, currently supports English, uses a 7.7B backbone plus a 300M audio decoder, and ships under a modified MIT license. Self-hosting removes the per-minute model fee and keeps runtime audio inside infrastructure you control, but the official repository recommends 24 GB VRAM for bf16 inference and estimates a 30 to 40 GB first download. The widely repeated 110 ms figure is Miso's hosted API time-to-first-byte on H100-class hardware, not a promise for local GPUs. Start with the API when volume is uncertain or speed to market matters. Self-host when measured steady volume, data control or deployment constraints outweigh GPU, redundancy, monitoring and engineering costs. Pilot both on the same consented voices and production scripts before committing. **Miso TTS is a credible open-weight option for expressive English voice applications, but self-hosting is not the automatic winner.** The model is large, the public 110 ms figure describes Miso's hosted API on H100-class hardware, and the downloadable inference path needs production engineering around it. Most early products should validate demand with the API. Teams with stable volume, strict data boundaries, or on-premises requirements should benchmark self-hosting on their own hardware. You may have seen the model called **Miso One**. Miso Labs' official repository and model card call it **Miso TTS 8B**. This guide uses the official name while answering the buying question behind both searches: should you pay by generated minute, operate the model yourself, or test a hybrid? ## Miso TTS 8B facts that change the deployment decision The useful facts are less viral than the demo clips. They determine whether the model fits your product and infrastructure. | Question | Published answer | What it means for a buyer | | --- | --- | --- | | Model size | About 8.2B parameters in total | This is not a lightweight CPU TTS engine. | | Language support | English only | A multilingual product needs another model or a routed voice stack. | | Local GPU target | 24 GB VRAM recommended for bf16 or fp16 | An RTX 3090, RTX 4090, A5000 or L4 class card is the documented starting point. | | Initial local download | Roughly 30 to 40 GB | Plan model distribution, cache warming and cold-start operations. | | Voice cloning | Prompted generation from audio context | You need the prompt transcript, consent controls and secure voice asset storage. | | Conversation mode | Individual turns, half-duplex | It does not solve turn-taking or full-duplex interruption by itself. | | License | Modified MIT | Commercial use is broad, with an attribution condition for exceptionally large products. | | Watermarking | Enabled by default | Production deployments should manage a private watermark key and preserve disclosure controls. | ## The 110 ms claim is API TTFB, not local end-to-end latency Miso publishes 110 ms as its hosted service latency. The project's own [GitHub documentation](https://github.com/MisoLabsAI/MisoTTS) adds the boundary that matters: public references to 110 ms describe hosted production API **time-to-first-byte on H100-class hardware**, not the unoptimized local inference path. It explicitly tells local users to expect materially slower startup and generation on consumer or workstation GPUs. Time-to-first-byte is also only one clock in a voice agent. The user experiences capture, speech recognition, end-of-turn detection, LLM generation, TTS queueing, first audio, playback buffering and interruption handling. Our [Nemotron 3.5 ASR production review](/blog/nvidia-nemotron-3-5-asr-production-review/) shows why an 80 ms audio chunk is not an 80 ms transcript. The same discipline applies here: a 110 ms first byte is not a 110 ms conversation. Benchmark at least P50 and P95 for request-to-first-audio, real-time factor, full-utterance completion and interruption recovery. Record cold and warm starts separately. A fast median with long cold starts can still be a poor production experience. ## What makes the speech expressive? Miso's architecture is inspired by Sesame CSM. A 7.7B transformer backbone processes interleaved text and audio context and predicts the first Mimi audio codebook index. A smaller 300M decoder predicts the remaining indices across 32 residual vector quantization codebooks. In plain language, the model can represent a much richer range of sound than a single flat audio-token vocabulary and can condition its next voice turn on prior audio, not just text. That explains the expressive pacing and tone in Miso's published samples. It does not independently prove that the model beats every hosted TTS system on your accents, domain names, numbers or noisy phone audio. Miso's [architecture announcement and audio samples](https://www.misolabs.ai/blog/miso-tts-8b) are first-party evidence. Treat them as a reason to test, not as a substitute for a blind listening evaluation. ## Miso TTS API vs self-hosting | Decision factor | Miso API | Self-hosted Miso TTS | | --- | --- | --- | | Time to first pilot | Usually lower integration effort | Environment, weights, GPU serving and operations come first | | Published latency | 110 ms TTFB target, H100-class hosted path | No equivalent local guarantee, measure your GPU and serving stack | | Cost shape | Subscription plus generated-minute overage | GPU, idle capacity, engineering, monitoring and redundancy | | Scaling | Provider absorbs most capacity management | Your team owns queues, autoscaling, failover and upgrades | | Runtime audio boundary | Audio and text pass through the provider service | Can remain inside infrastructure you control | | Customization | Hosted voices and plan-dependent custom voice features | Full control of inference code, weights and surrounding pipeline | | Best fit | Uncertain demand, rapid pilots, small teams | Stable volume, on-premises need, strict data boundary, capable platform team | The API is not inherently less serious and self-hosting is not inherently more private. Privacy depends on contracts, retention, logging and architecture. Self-hosting gives you control of the runtime boundary, then makes you responsible for every copy of prompt audio, cloned voice assets, logs, backups and generated files. ## What does the Miso API cost? As of 26 July 2026, Miso's [published monthly pricing](https://www.misolabs.ai/pricing) lists Starter at $20 for 150 included minutes plus $0.15 per extra minute, and Scale at $100 for 1,000 included minutes plus $0.10 per extra minute. Enterprise lists volume pricing below $0.08 per minute and on-premises deployment as a custom option. Prices and plan details can change, so recalculate from the live pricing page before procurement. Do not compare that per-minute price with only the electricity consumed by a local GPU. Use total monthly cost: `self-hosted TCO = GPU or server + idle headroom + storage and egress + observability + redundancy + security + engineering and on-call` `API TCO = plan + generated-minute overage + integration + vendor-risk controls` The break-even question is: `break-even minutes = monthly self-hosted TCO / effective API cost per generated minute` Use utilization, not wishful capacity. A GPU reserved for low and bursty traffic can be more expensive than an API even when its marginal generation cost looks tiny. Conversely, a fully utilized existing GPU fleet can make self-hosting attractive earlier. Our broader [local model versus API break-even framework](/blog/local-models-vs-apis-break-even-eu-2026/) covers the same utilization trap. ## Is the model really free and open source? The code and weights are publicly downloadable and there is no per-minute model license fee for ordinary self-hosting. The precise label is still important. Miso TTS uses a [modified MIT license](https://huggingface.co/MisoLabs/MisoTTS/blob/main/LICENSE). It grants broad rights to use, modify and sell the software. If a product using it exceeds 50 million monthly active users or $10 million in monthly revenue, the license requires prominent “Miso Labs” attribution in the product interface. That condition will not affect most teams, but “MIT” and “modified MIT” are not interchangeable in a due-diligence checklist. Record the model version, license text and dependency licenses when you approve the stack. This article is a technical procurement guide, not legal advice. ## Miso TTS is a voice component, not a complete voice agent The released model generates speech. It does not supply production speech recognition, turn detection, interruption handling, dialogue policy, tool execution, abuse prevention or telephony. Miso's own research note says the current system models individual turns and half-duplex audio, while turn-taking and full-duplex conversation remain future work. A production voice agent still needs: - streaming ASR and end-of-turn detection; - an LLM or dialogue system with tool and policy controls; - TTS serving with queues, cancellation and fallback voices; - barge-in and interruption behavior; - consent, retention and voice-asset access controls; - quality, latency, abuse and cost monitoring. If the product must speak more than English, Miso TTS 8B cannot be the only voice path today. Route languages to tested alternatives or postpone the model decision until multilingual quality is proven. ## Voice cloning changes the risk model The repository demonstrates prompted generation by passing audio and its transcript as context. That is technically useful and operationally sensitive. A ten-second recording is not just another media file when it can drive a reusable synthetic identity. Require explicit permission for every cloned voice, bind voice assets to an owner and allowed purpose, restrict export, log generation, rotate watermark keys, and provide a fast revocation path. Generated audio is watermarked by default in the public code. Keep watermarking enabled unless a documented requirement and risk review justify a change. For EU deployment, Article 50 transparency duties become applicable from 2 August 2026 for covered systems. The European Commission's [transparency guidance and Code of Practice](https://digital-strategy.ec.europa.eu/en/faqs/code-practice-transparency-ai-generated-content) address machine-readable marking of generated audio, disclosure for deepfakes, and informing people when they interact with AI. Scope and exceptions depend on the use case. Use our [EU AI Act Article 50 implementation checklist](/blog/eu-ai-act-article-50-checklist/) as an engineering starting point and obtain legal review for the final interpretation. ## Who should choose each deployment path? | Your situation | Recommended first move | Reason | | --- | --- | --- | | Prototype with unknown usage | API | Learn product demand before owning GPU operations. | | English voice agent with a hard launch date | API, then benchmark self-hosting in parallel | Protect delivery while collecting a real workload. | | Steady high volume and an existing GPU platform team | Self-hosted pilot | Utilization and operational capability can support the economics. | | Audio must stay inside a controlled environment | Self-hosted or negotiated on-premises | The runtime data boundary is a product requirement. | | Multilingual product | Evaluate another or routed model stack | The released Miso TTS 8B supports English only. | | Full-duplex conversational agent | Architecture pilot, not a model-only decision | Miso supplies TTS turns, not the full interruption and turn-taking system. | ## A two-week evaluation that produces a buying decision 1. **Freeze the test set.** Use consented voices and 100 to 300 real scripts covering names, numbers, abbreviations, emotions, long turns and difficult pronunciation. 2. **Run the same corpus through API and local paths.** Keep prompts, voice context and output settings identical. 3. **Measure the full latency curve.** Capture cold and warm P50, P95 and P99 request-to-first-audio, real-time factor and completed-utterance time. 4. **Blind-rate quality.** Score naturalness, speaker similarity, pronunciation, emotional fit and consistency without telling reviewers which path generated the clip. 5. **Load-test failure behavior.** Test concurrency, queue growth, cancellation, GPU loss, provider failure and fallback voice quality. 6. **Price production, not the demo.** Model three realistic volume bands and include idle headroom, redundancy and engineering. 7. **Review consent and disclosure.** Confirm who may clone which voice, where assets live, how they are revoked and how AI audio is marked. Define pass and kill thresholds before the test. That keeps a striking demo from overruling the operating model. If you need a bounded commercial format, our [paid pilot versus proof-of-concept guide](/blog/paid-pilot-vs-poc-vs-design-partner/) shows how to turn technical uncertainty into an explicit decision. ## Sources and claim boundaries Model size, architecture, English-only support, prompted generation, local requirements, download size, hosted-latency boundary and default watermarking come from the [MisoTTS repository](https://github.com/MisoLabsAI/MisoTTS). The RVQ design, 7.7B backbone, 300M decoder and half-duplex limitation come from Miso Labs' [release note](https://www.misolabs.ai/blog/miso-tts-8b). API prices and features come from [Miso's pricing page](https://www.misolabs.ai/pricing). License wording comes from the [published model license](https://huggingface.co/MisoLabs/MisoTTS/blob/main/LICENSE). EU dates and transparency measures come from the European Commission. Facts were checked on 26 July 2026. Wavect did not independently reproduce Miso's audio quality or 110 ms API claim. ## Frequently Asked Questions ### What is Miso One? Miso One is the product-facing name commonly used for Miso Labs' voice offering. The official open-weight project is named Miso TTS 8B. It is an English text-to-speech model that can condition on prior audio for expressive speech and voice cloning. ### Can I self-host Miso TTS? Yes. Miso publishes inference code and weights. The official repository recommends 24 GB VRAM for bf16 or fp16 interactive use and estimates a 30 to 40 GB initial download. CPU inference is possible but described as slow. ### Does Miso TTS really have 110 ms latency? Miso publishes 110 ms for its hosted API. Its repository clarifies that this is time-to-first-byte on H100-class hardware, not a guarantee for the downloadable local inference path or the full voice-agent response. ### How much does the Miso TTS API cost? On 26 July 2026, published monthly pricing listed Starter at $20 with 150 minutes and $0.15 per extra minute, and Scale at $100 with 1,000 minutes and $0.10 per extra minute. Enterprise pricing was custom. Check the live page because prices can change. ### Is Miso TTS free for commercial use? The model has no ordinary per-minute license fee and uses a modified MIT license with broad commercial rights. Products above 50 million monthly active users or $10 million monthly revenue must prominently display Miso Labs attribution. Review the current license with counsel. ### Does Miso TTS support languages other than English? The official repository currently lists English only. A multilingual application needs another model, a language-routed stack or a later Miso release that passes your own quality tests. ### Should I use the Miso API or self-host? Use the API first when demand is uncertain, the team is small or launch speed matters. Benchmark self-hosting when volume is steady, runtime audio must stay inside controlled infrastructure, or you already operate GPU services. Compare full TCO and production latency on the same workload. ## Final thoughts Miso TTS is interesting because it puts expressive, audio-conditioned speech behind both downloadable weights and a managed API. That gives teams a real deployment choice. It does not erase the work around the model. The practical answer is usually staged. Validate product demand and voice quality with the fastest reversible path. At the same time, collect the volume, latency, consent and failure data needed to price self-hosting honestly. Move the model in-house only when the measured operating case beats the managed service or the data boundary makes the decision for you. Keep the headline precise: 110 ms is hosted API time-to-first-byte on H100-class hardware. Keep the product decision broader: users experience the entire voice loop, finance pays for the entire operating stack, and risk owns every cloned voice. ## You may also like.. [**Nemotron 3.5 ASR for voice agents** Evaluate the speech-to-text half of a self-hosted voice stack, including multilingual quality, streaming latency and GPU economics.](/blog/nvidia-nemotron-3-5-asr-production-review/) [**Local models versus APIs** Use a full-cost break-even model before moving AI inference onto hardware you operate.](/blog/local-models-vs-apis-break-even-eu-2026/) Models and infrastructure ## Continue through this cluster Model selection, inference economics, local deployment, compression and serving architecture. [Start with the cornerstone**Self-Hosting LLMs in the EU: When Open Weights Actually Pay Off**](/blog/self-hosting-llms-eu-cost/) - [Transformers.js Browser AI: When Local Inference Belongs in Your Product](/blog/transformers-js-browser-ai-guide/) - [How to Self-Host LiteLLM in Production: 2026 Guide](/blog/self-host-litellm-production-2026/) - [AI-Ready Company Wiki: Architecture and Build Guide](/blog/ai-ready-company-wiki/) - [Does Claude Watermark Text? The 2026 API Answer](/blog/claude-text-watermark-api-2026/) - [OpenKB Review: Knowledge Compiler vs RAG](/blog/openkb-review-vs-rag/) Inbox, without the noise ## Follow the work that matters to you Get a short email when we publish something new. Follow the whole blog or only the problems you care about. [**Back**](/blog/overview/) [![Kevin Riedl](/img/team/kevin.webp)](/team/kevin-riedl/) [Kevin Riedl](/team/kevin-riedl/) https://linkedin.com/in/wsdt 11 min read · 26 Jul 2026 Last reviewed July 26, 2026 [**Next**](/blog/nvidia-nemotron-3-5-asr-production-review/) New posts by email × × Get new posts by email A short email when we publish. Free, no tracking. ## Structured Data ```json { "@context": "https://schema.org", "@graph": [ { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ], "employee": [ { "@id": "https://wavect.io/team/kevin-riedl/#person", "@type": "Person", "jobTitle": "Managing Director", "name": "Kevin Riedl", "url": "https://wavect.io/team/kevin-riedl/", "worksFor": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] } }, { "@id": "https://wavect.io/team/christof-jori/#person", "@type": "Person", "jobTitle": "Managing Director", "name": "Christof Jori", "url": "https://wavect.io/team/christof-jori/", "worksFor": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] } } ], "founder": [ { "@id": "https://wavect.io/team/kevin-riedl/#person", "@type": "Person", "jobTitle": "Managing Director", "name": "Kevin Riedl", "url": "https://wavect.io/team/kevin-riedl/", "worksFor": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] } }, { "@id": "https://wavect.io/team/christof-jori/#person", "@type": "Person", "jobTitle": "Managing Director", "name": "Christof Jori", "url": "https://wavect.io/team/christof-jori/", "worksFor": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] } } ], "legalRepresentative": [ { "@id": "https://wavect.io/team/kevin-riedl/#person", "@type": "Person", "jobTitle": "Managing Director", "name": "Kevin Riedl", "url": "https://wavect.io/team/kevin-riedl/", "worksFor": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] } }, { "@id": "https://wavect.io/team/christof-jori/#person", "@type": "Person", "jobTitle": "Managing Director", "name": "Christof Jori", "url": "https://wavect.io/team/christof-jori/", "worksFor": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] } } ], "name": "Wavect GmbH", "subjectOf": { "@id": "https://wavect.io/verified-claims.json#dataset", "@type": "Dataset", "creator": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] }, "description": "A machine-readable registry of quantitative and qualitative claims published by Wavect, with review dates, localized page appearances and public third-party citations where available.", "inLanguage": "en", "isAccessibleForFree": true, "license": "https://creativecommons.org/licenses/by/4.0/", "name": "Wavect verified publication claims", "url": "https://wavect.io/verified-claims.json" }, "url": "https://wavect.io/" }, { "@id": "https://wavect.io/team/kevin-riedl/#person", "@type": "Person", "jobTitle": "Managing Director", "name": "Kevin Riedl", "sameAs": [ "https://www.wikidata.org/wiki/Q139796365", "https://www.linkedin.com/in/wsdt", "https://github.com/wsdt" ], "url": "https://wavect.io/team/kevin-riedl/", "worksFor": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] } }, { "@id": "https://wavect.io/team/christof-jori/#person", "@type": "Person", "jobTitle": "Managing Director", "name": "Christof Jori", "sameAs": [ "https://www.wikidata.org/wiki/Q139796367", "https://www.linkedin.com/in/jocr77/", "https://github.com/jo-chris" ], "url": "https://wavect.io/team/christof-jori/", "worksFor": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] } }, { "@id": "https://wavect.io/#website", "@type": "WebSite", "inLanguage": [ "en", "de", "es", "zh" ], "name": "Wavect", "potentialAction": { "@type": "SearchAction", "query-input": "required name=search_term_string", "target": { "@type": "EntryPoint", "urlTemplate": "https://wavect.io/search/?q={search_term_string}" } }, "publisher": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] }, "url": "https://wavect.io/" }, { "@id": "https://wavect.io/blog/miso-tts-self-hosted-vs-api/#webpage", "@type": "WebPage", "dateModified": "2026-07-26", "inLanguage": "en", "isPartOf": { "@id": "https://wavect.io/#website", "@type": "WebSite" }, "lastReviewed": "2026-07-26", "url": "https://wavect.io/blog/miso-tts-self-hosted-vs-api/" } ] } ``` ```json { "@context": "https://schema.org", "@type": "BlogPosting", "abstract": "Miso One is the product-facing name people use for Miso Labs' Miso TTS 8B. The downloadable model can condition on short prompt audio for voice cloning, currently supports English, uses a 7.7B backbone plus a 300M audio decoder, and ships under a modified MIT license. Self-hosting removes the per-minute model fee and keeps runtime audio inside infrastructure you control, but the official repository recommends 24 GB VRAM for bf16 inference and estimates a 30 to 40 GB first download. The widely repeated 110 ms figure is Miso's hosted API time-to-first-byte on H100-class hardware, not a promise for local GPUs. Start with the API when volume is uncertain or speed to market matters. Self-host when measured steady volume, data control or deployment constraints outweigh GPU, redundancy, monitoring and engineering costs. Pilot both on the same consented voices and production scripts before committing.", "articleBody": " Blog overview/AI and agents/Models and infrastructure Miso TTS Self-Hosted vs API: Cost, Latency and VRAM Reality TL;DR Miso One is the product-facing name people use for Miso Labs' Miso TTS 8B. The downloadable model can condition on short prompt audio for voice cloning, currently supports English, uses a 7.7B backbone plus a 300M audio decoder, and ships under a modified MIT license. Self-hosting removes the per-minute model fee and keeps runtime audio inside infrastructure you control, but the official repository recommends 24 GB VRAM for bf16 inference and estimates a 30 to 40 GB first download. The widely repeated 110 ms figure is Miso's hosted API time-to-first-byte on H100-class hardware, not a promise for local GPUs. Start with the API when volume is uncertain or speed to market matters. Self-host when measured steady volume, data control or deployment constraints outweigh GPU, redundancy, monitoring and engineering costs. Pilot both on the same consented voices and production scripts before committing. Miso TTS is a credible open-weight option for expressive English voice applications, but self-hosting is not the automatic winner. The model is large, the public 110 ms figure describes Miso's hosted API on H100-class hardware, and the downloadable inference path needs production engineering around it. Most early products should validate demand with the API. Teams with stable volume, strict data boundaries, or on-premises requirements should benchmark self-hosting on their own hardware. You may have seen the model called Miso One. Miso Labs' official repository and model card call it Miso TTS 8B. This guide uses the official name while answering the buying question behind both searches: should you pay by generated minute, operate the model yourself, or test a hybrid? Miso TTS 8B facts that change the deployment decision The useful facts are less viral than the demo clips. They determine whether the model fits your product and infrastructure. Miso TTS 8B deployment facts, checked 26 July 2026 QuestionPublished answerWhat it means for a buyer Model sizeAbout 8.2B parameters in totalThis is not a lightweight CPU TTS engine. Language supportEnglish onlyA multilingual product needs another model or a routed voice stack. Local GPU target24 GB VRAM recommended for bf16 or fp16An RTX 3090, RTX 4090, A5000 or L4 class card is the documented starting point. Initial local downloadRoughly 30 to 40 GBPlan model distribution, cache warming and cold-start operations. Voice cloningPrompted generation from audio contextYou need the prompt transcript, consent controls and secure voice asset storage. Conversation modeIndividual turns, half-duplexIt does not solve turn-taking or full-duplex interruption by itself. LicenseModified MITCommercial use is broad, with an attribution condition for exceptionally large products. WatermarkingEnabled by defaultProduction deployments should manage a private watermark key and preserve disclosure controls. The 110 ms claim is API TTFB, not local end-to-end latency Miso publishes 110 ms as its hosted service latency. The project's own GitHub documentation adds the boundary that matters: public references to 110 ms describe hosted production API time-to-first-byte on H100-class hardware, not the unoptimized local inference path. It explicitly tells local users to expect materially slower startup and generation on consumer or workstation GPUs. Time-to-first-byte is also only one clock in a voice agent. The user experiences capture, speech recognition, end-of-turn detection, LLM generation, TTS queueing, first audio, playback buffering and interruption handling. Our Nemotron 3.5 ASR production review shows why an 80 ms audio chunk is not an 80 ms transcript. The same discipline applies here: a 110 ms first byte is not a 110 ms conversation. Benchmark at least P50 and P95 for request-to-first-audio, real-time factor, full-utterance completion and interruption recovery. Record cold and warm starts separately. A fast median with long cold starts can still be a poor production experience. What makes the speech expressive? Miso's architecture is inspired by Sesame CSM. A 7.7B transformer backbone processes interleaved text and audio context and predicts the first Mimi audio codebook index. A smaller 300M decoder predicts the remaining indices across 32 residual vector quantization codebooks. In plain language, the model can represent a much richer range of sound than a single flat audio-token vocabulary and can condition its next voice turn on prior audio, not just text. That explains the expressive pacing and tone in Miso's published samples. It does not independently prove that the model beats every hosted TTS system on your accents, domain names, numbers or noisy phone audio. Miso's architecture announcement and audio samples are first-party evidence. Treat them as a reason to test, not as a substitute for a blind listening evaluation. Miso TTS API vs self-hosting Decision factorMiso", "articleSection": "Engineering", "author": { "@id": "https://wavect.io/team/kevin-riedl/#person", "@type": "Person", "name": "Kevin Riedl", "sameAs": [ "https://www.wikidata.org/wiki/Q139796365", "https://www.linkedin.com/in/wsdt", "https://github.com/wsdt" ], "url": "https://wavect.io/team/kevin-riedl/" }, "dateModified": "2026-07-26", "datePublished": "2026-07-26", "description": "Miso One is the product-facing name people use for Miso Labs' Miso TTS 8B. The downloadable model can condition on short prompt audio for voice cloning, currently supports English, uses a 7.7B backbone plus a 300M audio decoder, and ships under a modified MIT license. Self-hosting removes the per-minute model fee and keeps runtime audio inside infrastructure you control, but the official repository recommends 24 GB VRAM for bf16 inference and estimates a 30 to 40 GB first download. The widely repeated 110 ms figure is Miso's hosted API time-to-first-byte on H100-class hardware, not a promise for local GPUs. Start with the API when volume is uncertain or speed to market matters. Self-host when measured steady volume, data control or deployment constraints outweigh GPU, redundancy, monitoring and engineering costs. Pilot both on the same consented voices and production scripts before committing.", "headline": "Miso TTS Self-Hosted vs API: Cost, Latency and VRAM Reality", "image": "https://wavect.io/img/blog/headers/header_miso-tts-self-hosted-vs-api.svg", "inLanguage": "en", "keywords": "Miso TTS, Voice AI", "mainEntityOfPage": { "@id": "https://wavect.io/blog/miso-tts-self-hosted-vs-api/", "@type": "WebPage" }, "publisher": { "@id": "https://wavect.io/#organization", "@type": [ "Organization", "ProfessionalService", "LocalBusiness" ] }, "url": "https://wavect.io/blog/miso-tts-self-hosted-vs-api/", "wordCount": 2614 } ``` ```json { "@context": "https://schema.org", "@type": "BreadcrumbList", "itemListElement": [ { "@type": "ListItem", "item": "https://wavect.io/", "name": "Home", "position": 1 }, { "@type": "ListItem", "item": "https://wavect.io/blog/overview/", "name": "Blog overview", "position": 2 }, { "@type": "ListItem", "item": "https://wavect.io/blog/topics/ai-agents/", "name": "AI and agents", "position": 3 }, { "@type": "ListItem", "item": "https://wavect.io/blog/clusters/models-infrastructure/", "name": "Models and infrastructure", "position": 4 }, { "@type": "ListItem", "item": "https://wavect.io/blog/miso-tts-self-hosted-vs-api/", "name": "Miso TTS Self-Hosted vs API: Cost and Latency | ", "position": 5 } ] } ``` ```json { "@context": "https://schema.org", "@type": "FAQPage", "mainEntity": [ { "@type": "Question", "acceptedAnswer": { "@type": "Answer", "text": "Miso One is the product-facing name commonly used for Miso Labs' voice offering. The official open-weight project is named Miso TTS 8B. It is an English text-to-speech model that can condition on prior audio for expressive speech and voice cloning." }, "name": "What is Miso One?" }, { "@type": "Question", "acceptedAnswer": { "@type": "Answer", "text": "Yes. Miso publishes inference code and weights. The official repository recommends 24 GB VRAM for bf16 or fp16 interactive use and estimates a 30 to 40 GB initial download. CPU inference is possible but described as slow." }, "name": "Can I self-host Miso TTS?" }, { "@type": "Question", "acceptedAnswer": { "@type": "Answer", "text": "Miso publishes 110 ms for its hosted API. Its repository clarifies that this is time-to-first-byte on H100-class hardware, not a guarantee for the downloadable local inference path or the full voice-agent response." }, "name": "Does Miso TTS really have 110 ms latency?" }, { "@type": "Question", "acceptedAnswer": { "@type": "Answer", "text": "On 26 July 2026, published monthly pricing listed Starter at $20 with 150 minutes and $0.15 per extra minute, and Scale at $100 with 1,000 minutes and $0.10 per extra minute. Enterprise pricing was custom. Check the live page because prices can change." }, "name": "How much does the Miso TTS API cost?" }, { "@type": "Question", "acceptedAnswer": { "@type": "Answer", "text": "The model has no ordinary per-minute license fee and uses a modified MIT license with broad commercial rights. Products above 50 million monthly active users or $10 million monthly revenue must prominently display Miso Labs attribution. Review the current license with counsel." }, "name": "Is Miso TTS free for commercial use?" }, { "@type": "Question", "acceptedAnswer": { "@type": "Answer", "text": "The official repository currently lists English only. A multilingual application needs another model, a language-routed stack or a later Miso release that passes your own quality tests." }, "name": "Does Miso TTS support languages other than English?" }, { "@type": "Question", "acceptedAnswer": { "@type": "Answer", "text": "Use the API first when demand is uncertain, the team is small or launch speed matters. Benchmark self-hosting when volume is steady, runtime audio must stay inside controlled infrastructure, or you already operate GPU services. Compare full TCO and production latency on the same workload." }, "name": "Should I use the Miso API or self-host?" } ] } ```