Native Vietnamese speaker to review AI-generated beginner listening audio (one 2.5-hour sitting)

We are building a Vietnamese listening podcast for absolute beginners — slow, clear speech that a learner understands from context, with no English translation. The narration is AI-generated, and we say so openly in every episode. Before we publish anything, we want a native Vietnamese speaker to tell us whether the audio is actually good enough to learn from. Machine testing already tells us it is intelligible and gets the tones approximately right. A machine cannot tell us whether a voice sounds like a real person, or whether an accent sounds authentically native rather than a convincing imitation. That is the entire reason we are paying for a native ear. WHAT THE JOB IS One sitting, about 2 to 2.5 hours of focused listening and writing. We pay US$150 fixed for it, which is deliberately above the hourly value of the time — we would rather you not feel rushed. - Five short clips, about 30 seconds each. All five read the exact same text in different voices, so you are comparing voice, tone and accent, not content. - Two full episodes, about 7 minutes each. These two also read the same script as each other. They answer what 30 seconds cannot: can a beginner follow one the whole way through, and does the voice hold together? - You fill in one short structured sheet per sample (checkboxes plus free text), then one comparison sheet at the end. We supply all the sheets as simple text files. You can write your answers in English or Vietnamese, whichever you think in. Everything is delivered as a download link. No software to install, no calls required, though we are happy to answer questions in Upwork messages before you start. WHAT WE ARE ACTUALLY ASKING YOU TO JUDGE The samples are blind and unlabeled. We deliberately will not tell you anything about the voices — where they are from, how they were made, or which one we prefer — because anything we tell you would contaminate your reaction. We are recruiting for one specific region of Vietnam and we are deliberately not saying which, because we do not want the audio judged against an expectation we planted. The questions that matter most to us: does this sound like a real person or like a machine; are the six tones right; does the accent sound genuinely native or imitated; and would you be comfortable with a student training their ear on this for months. PLEASE BE BLUNT — "NO" IS A PAID ANSWER Your review decides whether this project goes ahead at all. If the audio is not good enough for someone to learn from, the right outcome is that we stop, and telling us so is the single most useful thing you can do. It will not cost you the payment and it is not a disappointment to us. We would much rather hear an awkward truth now than after publishing a hundred episodes that teach people wrong Vietnamese. Please do not soften it. WHO WE ARE LOOKING FOR - Native Vietnamese speaker, grew up speaking it. - Strongly preferred: you grew up in one region and speak with that region's accent, rather than being comfortable across all three. A speaker fluent across dialects is more tolerant of a blended accent, which is exactly the defect we are trying to catch. - Experience teaching or tutoring Vietnamese to foreigners is a real plus — you will have heard learners mishear things and will know what actually trips them up. - Careful, specific written feedback. "Wrong tone at 1:42, the huyền on 'chào' sounds flat" is worth ten times "sounds pretty good". IN YOUR PROPOSAL, PLEASE TELL US 1. Where in Vietnam you grew up and what accent you speak with. 2. Whether you have taught Vietnamese to non-native learners, and for how long. 3. When you could do a single uninterrupted 2.5-hour sitting (your local time is fine). You do not need to write a long pitch. Three honest sentences beat a template. ONGOING WORK If this goes well we expect to want the same review on each new batch of episodes, on the same terms. We would rather build a relationship with one reviewer whose ear we trust than rotate through people. Posted by demades.ai.