Mandarin transcription buyer's guide: what to confirm before sending recordings

Why this matters

Mandarin recordings used for research, consulting, or strategic decision-making rarely need just a literal transcript. They need a document that analysts, partners, and decision-makers can reference repeatedly. In most projects, the most useful scoping question is not "is this an interview, meeting, or briefing" — it is "where is the real risk": mixed Mandarin-English, terminology density, speaker attribution, number sensitivity, confidentiality, or delivery pressure. This guide is organized around those risks so that first-time and recurring buyers can arrive with a clear scope map before sending the first recording.

Risk profile and service scope

A recording can be an interview, earnings call, forum, internal discussion, or briefing, but the handling requirements often repeat. Scope the project by the quality risk in the material instead of the meeting format. The four risk axes below frequently appear together — try to give a rough weighting at the first scoping conversation:

• Mandarin-English mixed transcription — bilingual recognition, acronym handling, and readable mixed-language formatting.
• Terminology-heavy transcription — finance, healthcare, manufacturing, semiconductor, product, or technical terms need subject-aware correction.
• Speaker-labeled and number-critical transcription — who said what, exact figures, units, dates, and Q&A structure need extra review.
• Confidential offline manual transcription — tool use, access, storage, and deletion rules are part of the scope.

Language scope and mixed English

Mandarin Chinese is the most controllable delivery scope at FingerPower, with Simplified Chinese as the default deliverable. Mixed English is common in research and business recordings. The difficulty is not merely whether English should be kept; it is recognizing English terms, acronyms, and brand names accurately before deciding how to format them. Once the recognition is wrong, no amount of polish will recover it — language scope is fundamentally a recognition problem first, a formatting problem second.

Terminology density

Dense terminology raises recognition risk. A sentence that a general listener treats as "a few unusual words" can become several entirely unrelated words in an ASR draft, pushing the overall reading of the sentence in the wrong direction. Before sending recordings, gather available references so the reviewer has a vocabulary map before listening — five minutes of preparation up front often saves more than thirty minutes of rework downstream.

Accuracy expectations: speakers, numbers, and uncertainty markers

Some projects need more than "readable paragraphs". They need named speakers, clear Q&A structure, exact figures, percentages, units, and dates, plus uncertainty markers where the audio cannot support a confident determination. Confirm whether these checks matter before work starts — retrofitting after delivery typically costs several times more than confirming up front.

• Named speakers or role labels — for example "Zhang San / CFO" or "Interviewee A".
• Q&A structure — whether questions and answers must be visually separated for easier retrieval.
• Numbers and units — amounts, percentages, currencies, and time ranges checked back against the audio one by one.
• Timestamp granularity — every paragraph, every minute, or only at speaker changes.
• Uncertainty marker convention — for example "[?]" or "[inaudible]" for points that require client confirmation.

Confidentiality and data handling

Confidentiality is not just a promise on a website, and it is not just signing an NDA. What actually shifts risk is a set of auditable, concrete actions: who can open the file, how it is transferred, how long it is retained, whether deletion confirmation is provided, and whether cloud ASR or AI correction tools are excluded. An NDA is the floor, not the ceiling — before transferring any sensitive recording, get the five points below into writing.

• Access scope — which named individuals can open the file.
• Transfer method — encrypted channel or controlled shared drive.
• Retention window — 30 / 60 / 90 days, or a project-specific agreement.
• Deletion confirmation — written confirmation after the retention window.
• Tool exclusion — whether generic cloud ASR and AI correction are prohibited.

Turnaround expectations

A reasonable standard turnaround for a single recording is 48 hours, covering transcription, review, and final delivery. Express options are possible but should be scoped case by case — they typically depend on recording length, audio quality, terminology familiarity, and current workload. For recurring engagements, a stable cadence is usually more valuable than the fastest possible single turnaround.

Pricing models

Common pricing models for human Mandarin transcription fall into three forms, each suited to a different engagement shape. Ideal pricing reflects recording difficulty — audio quality, mixed English ratio, technical density, and confidentiality requirements — not just length.

• Per audio hour — the most common reference unit, easy for benchmarking, suited to ad-hoc projects with similar risk profiles.
• Per finished word — fits projects that need editorial polishing or structural cleanup, avoiding the perverse incentive of "longer recordings cost more".
• Per-project retainer / monthly engagement — fits recordings that arrive on a steady rhythm, and trades a higher commitment for cadence and terminology continuity.

How to evaluate a sample

Before formal engagement, ask for an anonymized side-by-side: the raw ASR draft and the human-reviewed output. A good sample is not just "reads smoothly" — it should make the reviewer's handling logic visible at each risk axis. Focus on:

• Whether terminology errors were identified and corrected, not hidden in fluent prose.
• Whether mixed English is preserved as the speaker actually used it, and formatted readably.
• Whether numbers, units, and personal names were checked against the audio.
• Whether uncertain points are explicitly marked rather than silently filled in.
• Whether paragraph structure supports research retrieval and downstream quoting.

One-off vs long-term cooperation

One-off projects work for isolated recordings or urgent scoping. Recurring research, consulting, and decision-making workflows benefit more from long-term cooperation — terminology, speaker conventions, and format preferences carry across projects so each new transcript does not restart from zero. The deciding factor is usually consistency, not unit price.

Pre-engagement checklist

Before sending the first recording, confirm the items below at an executable level — not "we will take confidentiality seriously" but "deletion confirmation provided 30 days after delivery"; not "on-time delivery" but "48-hour turnaround, express handled case by case". Writing each item at that granularity prevents most post-delivery disputes.

• Risk profile (mixed-language / terminology-heavy / speaker-and-number / confidential-offline) and language scope (Mandarin only / with Traditional or translation as added scope).
• Readable transcript style, and any verbatim, oral-flavor, or formal-prose requirements.
• Speaker labels, number checks, timestamp granularity, and formatting preferences.
• Existing glossaries, company / product name lists, reference materials, or prior delivery samples.
• NDA, named access scope, tool exclusion, retention window, and deletion confirmation method.
• Expected turnaround and delivery format (DOCX / PDF / Markdown / client template).
• Whether summaries, translations, or bilingual side-by-side outputs are part of the deliverable.