Dấu - See your Vietnamese tones

The company appears to be a solo developer project named Dấu — See your Vietnamese tones, built as part of an OpenAI hackathon. The author states it is an open-source Vietnamese tone practice lab that uses signal processing (YIN) and LLMs (OpenAI models) for pitch grading and coaching. It allows users to record syllables, compare them with validated contours, and receive feedback on tone accuracy.

What changed: The project was submitted to the OpenAI 2026 hackathon, indicating a development phase focused on building a functional prototype. There is no evidence of prior commercial activity or product release beyond this submission.

The single most important open question: Is there any evidence that Dấu has achieved user adoption or traction beyond its author's own testing and development?

• The description states that Dấu is an open-source Vietnamese tone practice lab.
• It offers two core experiences:
  • Tone Shapes: Learners hear a reference, watch its contour, record a syllable, and see both pitch curves overlaid.
  • Dialogue Practice: Applies tone skills across four connected scenes with 26 turns and 13 learner replies.
• The system uses:
  • YIN algorithm for pitch extraction.
  • Dynamic Time Warping (DTW) for contour comparison.
  • OpenAI models for transcription, coaching, reference speech, and meaning artwork.
• It runs locally in the browser, without requiring an API key.
• The app bundles validated audio, artwork, and practice fixtures.
• Reference recordings must pass validation before becoming ground truth.

Inference: The product is described as a self-contained educational tool for Vietnamese tone learning, combining signal processing with LLM-based coaching. It is not a commercial SaaS offering but rather an open-source prototype.

• The author states that most apps give learners a red X without showing what their voice actually did.
• Dấu aims to show the pitch curve, explain the accidental meaning, and provide physical correction.
• It positions itself as a tool for Vietnamese tone practice, not general language learning.
• The tagline is: “Dấu makes Vietnamese tones visible. Record a word, compare your pitch with validated contours, see the meaning your tone created, and get the exact physical fix.”

Claim: Dấu is positioned as a visual and interactive tool for tone correction in Vietnamese.

• The description states that the author’s family speaks Vietnamese, and he was frustrated by tone miscommunication.
• The product is built for Vietnamese learners, particularly those who struggle with tone-based meaning distinctions (e.g., ma vs. má).
• It targets users who want to practice tone accuracy and receive immediate feedback.

Inference: The ICP appears to be language learners of Vietnamese, especially those using digital tools for pronunciation practice.

• No pricing information is provided.
• The product is described as open-source (MIT license).
• It works without an API key, suggesting no monetization layer at this stage.
• There is no mention of paid features, subscriptions, or commercial use cases.

Not evidenced: No evidence of a business model or pricing structure beyond the open-source prototype.

• The system uses:
  • YIN algorithm for pitch extraction.
  • Dynamic Time Warping (DTW) for contour comparison.
  • OpenAI models for transcription, coaching, and reference speech.
• It runs locally in the browser, removing network latency.
• A browser Web Worker handles pitch grading.
• Server-side components include:
  • FastAPI for model calls.
  • BotID, quotas, concurrency limits, and kill switches.
• Reference recordings must pass lexical, signal, and contour validation before being accepted.
• The project is built with chatgpt, codex, gpt-4o-transcribe, gpt-5.6-sol, gpt-image-2, gpt-realtime-2.1, gpt-realtime-2.1-mini.

Inference: The technical stack suggests a hybrid approach combining signal processing and LLMs for tone grading and coaching. It is designed to be self-contained, with no external dependencies beyond OpenAI models.

• The author reports:
  • 4 dialogue scenes and 52 validated recordings.
  • 91.7% held-out acoustic-family accuracy on the current reference corpus.
  • About $12 in total OpenAI build spend.
  • Two Southern reference targets withheld due to validation failure.
• It was submitted to an OpenAI hackathon, indicating a development phase.
• No evidence of:
  • Users, customers, or adoption.
  • Revenue or monetization.
  • Product release or public availability beyond the Devpost submission.

Not evidenced: No traction data, user metrics, or commercial adoption is provided.

• The description states that most apps give learners a red X without showing what their voice actually did.
• It does not name specific competitors.
• The product appears to be focused on tone accuracy, which is a niche within language learning tools.
• No mention of existing tools or platforms in this space.

Not evidenced: No competitive landscape or comparison with other tone-learning tools.

• The project is described as a hackathon submission — no evidence of commercial viability or long-term product strategy.
• It is open-source, which may limit monetization opportunities.
• The author reports that two Southern reference targets still fail validation gates, indicating incomplete coverage.
• There is no evidence of user feedback, testing, or adoption beyond the author’s own use case.
• The system relies heavily on OpenAI models, which could pose risks if pricing or access changes.

Inference: The project lacks commercial traction and may not have evolved into a scalable product.

1. What is the long-term vision for Dấu beyond this hackathon prototype?
2. Are there plans to monetize or scale the tool beyond open-source?
3. How many users or learners have tested the tool outside of your own use case?
4. What are the technical limitations or scalability concerns with the current architecture?
5. Have you considered how to expand beyond the current reference corpus (e.g., Southern dialects)?
6. Is there any plan for integrating feedback loops or adaptive learning?