Miso Labs

Name: Miso Labs
Brand: Miso Labs

AI Native

The most emotive foundation models for voice. Backed by Y Combinator (P26).

Est 2025|Raised $500K

LinkedIn 638|Website|Web traffic 11K|FTEs 3|HQ San Francisco, California, US

MarketsFoundation Models

Published Sep 27, 2026

01Common complaints

- Regulated environments with strict procurement requirements - If SOC 2 Type II/ISO 27001 evidence, DPAs, subprocessor lists, and a documented security posture are required up front, Miso Labs has limited public documentation in the sources reviewed. - Teams needing immediate production rollout - If access is gated (waitlist) and pricing/SLAs are not clearly published, production planning can be risky. (https://www.daltalabs.com/) - Multilingual deployments - The open model explicitly states English-only support at present. (https://github.com/MisoLabsAI/MisoTTS) - Low-infrastructure teams (no GPU budget) - The model is not positioned as lightweight; local inference can be slow on CPU and expects high-VRAM GPUs for interactive use. (https://github.com/MisoLabsAI/MisoTTS) - Buyers who primarily need business-ready packaging - If the core requirement is turnkey dashboards, access controls, team management, and predictable billing, established commercial vendors may be a better fit. - Current limitations acknowledged by the vendor - English-only support (explicit). (https://github.com/MisoLabsAI/MisoTTS) - Local inference latency and hardware requirements are significant. - The 110 ms latency figure cited in some places refers to hosted API time-to-first-byte on H100-class hardware, not to local inference. (https://github.com/MisoLabsAI/MisoTTS) - Features on the roadmap (not yet available) - No official roadmap with dates was found in accessible sources. - Scale limitations (at what volume/size does it break down?) - Not publicly quantified. - Likely scaling constraints include GPU concurrency, streaming performance, and caching strategy, which are typical for TTS systems. - Security or compliance gaps - Publicly visible SOC 2/ISO documentation was not found. - API limitations or integration constraints - Hosted API details are not visible on daltalabs.com in captured sources; waitlist suggests limited or staged access. (https://www.daltalabs.com/)

02Why they buy

- Competitive advantages validated by available artifacts (not only vendor claims) - Direct third-party user-review validation is limited. - Primary artifacts indicate: - An open-weight release with runnable code and explicit system requirement guidance. (https://github.com/MisoLabsAI/MisoTTS) - Unique features and capabilities competitors lack - An open-weight 8B model aimed at expressive conversational speech can be a differentiator compared with closed APIs. - Differentiation from competitors based on reviews - Not enough direct comparative reviews were found to draw robust conclusions. - Areas where competitors clearly have advantages - Clear enterprise packaging and security documentation (commonly available from larger vendors). - Established pricing models and more abundant third-party reviews.

03Field reports

Customer Testimonials

- Important caveat - The name “Miso Labs” is used by other companies/products (e.g., Miso.ai ecommerce search; Miso Robotics), which makes review discovery noisy. This report focuses on Miso Labs as the voice foundation model company tied to the vendor’s official website. - Review platforms (G2, etc.) - G2 reviews for “Miso.ai” appear to refer to a different product (ecommerce search/personalization) and are not attributable to the voice-model company. (G2 product pages and reviews) - No clearly attributable listings or reviews for the voice-model “Miso Labs” were found on TrustRadius during this session. (TrustRadius product listings) - No product page or reviews for the voice-model “Miso Labs” were located on Gartner Peer Insights in the accessible results. (Gartner Peer Insights search results) - Community / social discussion - Hacker News: some threads reference unrelated projects also named “Miso,” illustrating the name-collision risk in community searches. (Hacker News threads) - Third-party commentary: a Fluenta write-up about YC Spring 2026 references Miso in the context of category competition (ElevenLabs, Cartesia, OpenAI realtime API) and includes a secondary quote suggesting some users distinguish between voice similarity and emotional expression; this is secondary reporting and requires direct verification. (Third-party industry write-up / report) - What customers/users seem to value (limited evidence) - Availability of open weights and self-hosting options can be a meaningful adoption lever in developer communities. - The public GitHub repository provides detailed guidance about local performance expectations and hardware needs, which may reduce evaluation surprises. (Public GitHub repository and documentation) - Common criticisms and complaints (limited direct reviews) - Limited review coverage is a risk signal for enterprise buyers because it makes it harder to validate production maturity. - Secondary commentary suggests emotion quality may not meet all expectations for some listeners; this requires direct validation via benchmarks and listening tests. (Third-party industry write-ups)

Public case studies

No formal case studies (customer names + before/after metrics) were found on the vendor's website in the captured sources.

Vendor-published metrics and verification gaps

- Latency claims - Some third-party posts cite a specific low-latency figure, but those are not primary sources and may omit conditions or hardware context. (Third-party blog post) - The project's GitHub repository provides corrective framing: the cited latency figure refers to hosted API time-to-first-byte on high-end GPU hardware and does not apply to local inference, which highlights how latency claims can be misinterpreted. (Official project repository)

Mixed outcomes

No documented failed deployments or negative case studies were found in accessible sources.
Given early-stage signals (waitlist, limited enterprise documentation), product rollout should be treated as higher risk and planned with mitigations (pilot phase, fallback vendor, exit strategy).

04Best fit

Ideal Customer Profile

- Company sizes - Early-stage to mid-market product teams that can tolerate some vendor ambiguity: - Small-to-mid engineering organizations or teams with strong ML/infra capability, with willingness to run pilots. - Enterprises may be a fit if security/compliance requirements can be satisfied via private documentation. - Industries and verticals where adoption is strongest (likely) - Conversational AI / customer support tooling - Interactive media / games - Creator platforms and narration tooling - Education / language learning - Roles and personas: decision makers AND end users - Decision makers - Head of AI / VP Engineering / CTO - Product lead for voice agent experiences - Security/Compliance (blocker role) - End users - ML engineers, applied scientists - Backend engineers integrating streaming TTS - Voice UX designers evaluating prosody and emotional delivery - Pain points and challenges the product addresses - More expressive output for voice interactions. - Potentially lower latency in vendor-hosted deployments for real-time conversation. - Option to self-host via open weights for data sovereignty and customization. - Qualification criteria: what makes a good fit - Organizations able to run a structured bake-off (listening tests, latency benchmarks, MOS scoring) and that possess GPU resources. - Organizations that value open weights and have a clear plan for watermarking and safety. - English-only output is acceptable for the initial deployment (vendor open-source repository). - Deal disqualifiers: what makes a poor fit - Requirement for published SOC 2/ISO attestations before any pilot. - Need for multilingual TTS in the short term. - Inability to support high-VRAM GPU inference combined with lack of quick API access (vendor/third-party documentation).

05The pitch

What the product does (core value proposition in objective terms)

- Miso Labs (brand) / Dalta Labs (domain) presents itself as a provider of expressive foundation models for voice, aiming to support expressive text-to-speech (TTS) and voice-agent behaviors (e.g., laughter, whispering, nuanced delivery) via a REST API and a web studio. - Public artifacts indicate the core deliverable is a voice model family centered on a multi-billion-parameter TTS model (branded Miso TTS / Miso One), distributed as open weights plus (separately) a hosted experience with waitlist-driven access.

Target market and market positioning

- Primary target users appear to be: - AI voice-agent developers who require lower-latency, more expressive speech generation. - Developers and researchers who want to self-host or experiment with open-weight models. - Startups building consumer-facing voice experiences (assistants, interactive media, game NPCs, customer support, etc.). - Positioning emphasizes emotion fidelity and latency as differentiators compared with mainstream TTS vendors.

Get a comprehensive analysis of Miso Labs including market position, competitive landscape, adoption trends, and peer benchmarks.

Download full report →