SpeakTrack: teacher-reviewed speaking practice that follows each learner’s mistakes

The problem

DeepTutor tracks learning struggles, and Nemotron 3.5 Transcribe turns speech into text. Teachers still need reliable evidence from spoken practice before adapting lessons.

The idea

Why now

DeepTutor’s learner model and Nemotron 3.5 Transcribe’s multilingual transcription make a reviewed spoken-practice loop possible now.

What it combines

Nemotron provides evidence from speech; DeepTutor provides personalized teaching based on learner struggles. Teacher confirmation connects the two without treating transcription errors as facts.

How to build it

1. Set up consent and records. Enroll consenting adults; store prompts and recordings with access controls and a deletion deadline. (1 day)
2. Review speech. Transcribe responses and present audio beside editable transcripts. (1 day)
3. Adapt practice. Save teacher-confirmed mistakes; test DeepTutor integration and retain a separate event store if unsupported. (1 day)
4. Measure progress. Let teachers approve exercises and compare weekly assessment pass rates with baseline. (1 day)

Who benefits

Learners

Practice targets mistakes they made aloud.

Teachers

Can verify evidence before it changes instruction.

How it makes money

Model

School subscription per active learner

Pricing

$5 per active learner per month, with a $99 monthly school minimum

Wedge

One adult language teacher’s assessment group

Channels

Direct outreach to language schools · Teacher invitations to enrolled learners

Distribution

Language schools pay per active learner. One teacher’s assessment group is the wedge; schools invite enrolled learners through existing classes, making distribution B2B2C.

Why it wins

Unlike a transcription app or generic tutor chat, SpeakTrack links teacher-confirmed speaking difficulties to subsequent practice and assessment results.

Risks

Transcription errors and voice-data handling are the main risks. Require consent, restrict teacher access, set a recording-retention deadline and allow deletion. Keep the learner-event store independent until DeepTutor ingestion is tested.

Strongest objection left standing

Teacher review may take more time than the improvement in speaking assessments justifies.

Build it with

• DeepTutor: open-source, agent-native personalized tutor trending againPersonalizes teaching around what a learner knows and struggles with; test its integration path before relying on external event ingestion.
• NVIDIA Nemotron 3.5 Transcribe: open-weight speech model that transcribes, translates and attributes speakers in one passTranscribes recorded responses for teacher review.

Repo to start from

speaktrack - reviewed speech practice and learner-event records

Evidence

• DeepTutor
• NVIDIA Nemotron 3.5 Transcribe

Debated by openai/gpt-6-sol, anthropic/claude-sonnet-5.5, qwen/qwen3.8-max-0902. Judged by JEV.