AURA / DealTalk: AI Sales Voice Practice

01 · Project overview

Human roleplay is costly, scenarios are hard to reproduce and feedback can be delayed. AURA organizes personas, emotions, difficulty and sales situations into a mobile web practice experience for expression, discovery and objection handling. Customers here are AI training personas, not external client engagements.

AURA / DealTalk: AI Sales Voice Practice — screenshot 1

02 · My responsibilities

• Handled deployment and full-stack frontend integration of voice and practice flows.
• Built manager/HR views and personal history, relationship records and growth insights.
• Worked on browser microphone capture, conversation pacing, interruption and fallback experiences.

03 · Features and user journey

• Persona Lab selects AI characters by region, industry, role, mood and difficulty.
• Practice supports calls, conversations and callbacks, with coaching feedback in history.
• Relationship records and Growth views show relationship changes, score trends and skill profiles.
• Twelve progressive Quest tasks unlock through stars, turning free practice into structured goals.

AURA / DealTalk: AI Sales Voice Practice — screenshot 2

AURA / DealTalk: AI Sales Voice Practice — screenshot 3

AURA / DealTalk: AI Sales Voice Practice — screenshot 4

AURA / DealTalk: AI Sales Voice Practice — screenshot 5

04 · Voice pipeline and fallbacks

Microphone capture and VAD detect speech and pauses; qwen3-asr-flash transcribes, Qwen responds within scripted beats, and CosyVoice synthesizes speech. The flow supports interruption and next-utterance prefetch, with typed input when microphone or ASR fails. Health checks determine real-model availability.

05 · Outcomes and known limitations

The report records 14 concentrated commits, 39 base personas, 12 quests and eight TTS voices. Speaking is the implemented 50-point scoring component; the Thinking judge-crm endpoint is unimplemented, so the nominal 100-point framework is not a completed loop. Product knowledge is not yet wired into conversations, and some evaluation, verification and CRM endpoints remain placeholders.

AURA / DealTalk: AI Sales Voice Practice — screenshot 6

AURA / DealTalk: AI Sales Voice Practice — screenshot 7

06 · Reflection

A voice product needs more than speech: structured dialogue, timely feedback and continuity under failures matter. Rapid validation helps establish the flow, but testing and typing debt must be addressed, with working capabilities clearly distinguished from UI placeholders.

07 · Scope and source

This was a self-initiated or personally led practice during my Alibaba Cloud trainee period, supporting internal learning, collaboration or training rather than external client delivery. The account follows my project report dated October 10, 2026. Development and test figures describe that historical snapshot, not current production status. Screenshots come exclusively from my supplied folder; sensitive people and contact details are masked.