SpeakTranslation site

Speak Company Profile

Speak is a US-based AI founded in 2016language learningSpeak, based in San Francisco, was founded by 90-year-old entrepreneurs Connor Zwick and Andrew Hsu. The company initially focused on the Korean market, through AI technology to solve the "mute English" pain points, to provide personalized oral training services. 2022 access to OpenAI technology, Speak completed several rounds of financing, 2024 C round of financing of 78 million U.S. dollars, valuation of 1 billion U.S. dollars, become a unicorn in the field of AI education. Unicorn. As of 2025, its app downloads exceeded 10 million times, covering more than 40 countries around the world, and its annual recurring revenue (ARR) was close to $50 million, with an annual growth rate of more than 100%.

Speak Products & Services

1. core functionality
   • AI Speaking Tutor: Integrated speech recognition model Whisper and GPT-4 technology to provide real-time dialog practice. Users can conduct situational simulation dialog with AI (e.g., business negotiation, travel asking for directions), and the system provides instant feedback on pronunciation, grammar and vocabulary errors.
   • Personalized Courses: Generate customized lessons based on user level, interests and goals. For example, design relevant dialog scenarios for users preparing for a trip to Mexico.
   • Corporate ServicesLaunched "Speak for Business", a business conversation training program, with over 200 corporate customers and an employee adoption rate of 85%.
2. Technical Highlights
   • Live Roleplays: Real-time interaction between the user and the AI is realized in the form of a language phone call, and the system automatically adjusts sentence patterns and vocabulary according to proficiency to reduce response latency.
   • weakly supervised learning: Extracting data features by rules or pre-trained models to generate weakly supervised labels and reduce the reliance on manual labeling.
3. business model
   • C-Subscription: Paid access, $20 for a monthly subscription or $99 for an annual subscription, with a 7-day free trial available.
   • B-side cooperation: Customized language training solutions for corporate clients.

Speak Market Competitiveness

1. differentiated positioning
   • Speak focuses on the ability to "speak English", differentiating itself from game-based learning platforms such as Duolingo. Speak fills a gap in a market where 1.5 billion people around the world are trying to learn English, but speaking skills are generally weak.
2. technical barrier
   • Speech recognition optimization: It can accurately understand different accents and help users improve their pronunciation.
   • Data accumulation: Constructing large datasets of spoken examples labeled in a second language provides a unique advantage for model training.
3. Localized operations
   • Local teams have been established in markets such as Korea, Japan and Taiwan, with course programs and operational strategies tailored to local conditions. For example, the revenue share of the Korean market had exceeded 90%, and the revenue share of the Japanese market had reached 35%.
4. Capital and resources
   • Received investment from OpenAI Startup Fund, Accel, Khosla Ventures and other top organizations, with abundant capital to support technology development and market expansion.

Speak Development Prospects

1. market potential
   • Speak is expected to further expand its user base as global demand for language learning continues to grow and AI technology reduces the cost of speaking training. The company plans to expand to more countries and broaden the variety of source languages (e.g., support for Spanish, French, etc.).
2. Technology deepening
   • Exploring the fusion of self-supervision and weak supervision to generate pseudo-annotations using unlabeled data to reduce dependence on external knowledge.
   • Building an English fluency quantification system to support language proficiency testing.
3. ecological expansion
   • Expand B-side business, deepen cooperation with corporate clients and provide more comprehensive language training solutions.
   • Explore synthetic data technologies to address long-tail scene labeling challenges in areas such as autonomous driving (strengthened capabilities through acquisition of minority stake in Mindtech).
4. Challenges and responses
   • competitive pressure: Facing competition from giants such as Duolingo and Loora, it needs to maintain its advantage through technological differentiation.
   • data privacy: Enhance data security technology to meet compliance requirements in different regions of the world.