Voice and knowledge
How does Thai and English voice knowledge teach a trading assistant?
Use audio as context, verify transcription and source retrieval, and distinguish voice knowledge from live voice and trading consent.
Recorded voice knowledge captures your explanation and turns it into text for retrieval in a configured private workspace. It helps supply context; it is not a trading order or an accepted plan. TradeTwin's customer KB accepts audio only. Thai and English onboarding is available, while multilingual live conversational voice remains a separate, unavailable pilot feature.
Record one clear idea at a time
Explain the rule, the reason behind it, the conditions where it applies and the exceptions. State whether you are describing your normal method or an experiment. If you mix Thai and English market terms, make important numbers and units explicit. Avoid including API keys, passwords or another person's account information in a recording. The goal is useful context that you can inspect, not a long file whose meaning depends on unstated assumptions.
Inspect the transcript before relying on it
Speech recognition can misunderstand a symbol, decimal, negation or specialist term. A small difference can change the interpretation of a risk instruction. Compare the transcript with what you intended and correct the source before treating it as evidence. The customer pipeline has to finish transcription and indexed retrieval before the assistant can use it. Upload success alone does not prove that a usable passage was retrieved or that its meaning was preserved.
Know what the customer KB accepts
The private customer intake accepts recorded or uploaded audio, with configured limits of 95 MB and 60 minutes checked through a private stream probe. Video, document and text KB uploads, and URL imports, are refused. Private ASR and the transcription-provider configuration must be ready for the particular workspace. These limits describe the intake design, not a promise that an unprovisioned workspace can already transcribe a file or that unlimited voice processing is included.
Keep the source visible in the reasoning
A retrieved passage should be considered context from a particular recording, not proof that the statement is current market data. A recording about how you trade and a fresh account or price observation serve different purposes. If retrieval is unavailable, the assistant must make that limitation visible rather than claim to have checked your knowledge. When a new proposal relies on a passage, review whether it preserves the intended rule and separates AI suggestions from your instructions.
Distinguish live conversation from recorded knowledge
Recorded audio intake and a real-time spoken conversation are different products with different latency, provider and billing requirements. Live multilingual voice is being researched and is not active in this pilot. Thai and English text onboarding and voice-source explanations do not prove live voice quality or subscription-funded audio access. Any future live feature needs its own language testing. A transcribed instruction still requires a reviewed proposal and separate readiness checks before trading activation.
Questions worth asking
Does a recording automatically become a trading instruction?
No. It supplies knowledge context after processing and retrieval. The assistant may use it to help draft a plan, but the owner must review and accept the exact proposal, with execution readiness checked separately.
Can I use Thai with English trading terms?
You can explain your method that way, but inspect specialist terms and numbers in the transcript. Language support does not make every transcription error-free, and an uncertain passage should not silently become a risk rule.
Can I talk to the bot live right now?
Multilingual live conversational voice is not available in this pilot. Recorded audio knowledge and the private text interview are separate capabilities, subject to the customer's configured workspace and provider readiness.