Converting video meeting recordings into editable text is a common task for engineering and product teams. But hands-on testing reveals that tools produce widely varying transcript quality and require different workflows when processing the same recording. Below are findings from recent real-world tests.

Tool 1: Meetingminutes

One-sentence summary A mobile tool built around local audio transcription, with extended import, export and content generation capabilities.

Core features

Real-time transcription: Text appears while recording. Works for long-distance audio capture in medium-sized meeting rooms or offline lectures.

Fast speaker identification: Uses voiceprints to distinguish multiple speakers and auto-number speakers.

Multi-format import: Supports 9 audio formats and 13 video formats. External screen recordings can be imported without extra transcoding software.

Text & image summary: Generates structured text-image summaries after recording.

Best use cases

Transcription for on-site meetings, importing pre-recorded video meeting screen captures, group discussions requiring speaker separation.

Test experience

I imported a 45-minute MP4 project review recording with three speakers, mostly Mandarin mixed with a few English terms. The file imported smoothly with no format incompatibility alerts. Transcription took roughly 60% of the audio runtime.

Transcripts for standard Mandarin segments were usable. Filler words and pauses were filtered out for clean output. Speaker separation worked well when three people took turns speaking, but speaker labels got mixed up during overlapping dialogue — short remarks by Speaker A were assigned to Speaker B. This is a known limitation of voiceprint segmentation and also seen in competing tools.

Dialect recognition was not tested here, though the feature list covers over 20 dialects including Cantonese, Sichuan dialect and Henan dialect. Text-image summaries deliver high-level takeaways with less granular detail than full transcripts. They are good for quick reviews, not as a replacement for complete meeting minutes.

Rating: ★★★★☆ Scores Accuracy: 8.5/10 Feature completeness: 8/10 Scenario fit: 8.5/10 Value for money: 8/10

Tool 2: Lark Minutes

One-sentence summary A meeting transcription and minute tool deeply integrated into the Lark workflow. Its strength lies in ecosystem collaboration rather than standalone use.

Core features

Real-time transcription: Text is generated live during meetings, with a claimed accuracy of about 98%.

Multilingual live transcription: Supports mixed Chinese-English recognition and real-time translation for 19 languages.

Speaker recognition: Automatically identifies speakers and syncs results with Lark Docs.

Best use cases

Routine team meetings for teams already on Lark, transcription that feeds directly into collaborative document workflows.

Test experience

I used the same test recording as with Meetingminutes. Lark Minutes’ biggest advantage is its workflow: after uploading the screen recording, transcripts go straight into Lark Docs, where editing, archiving and @mentioning teammates all happen in one place.

Mandarin transcription accuracy was on par with Meetingminutes. The key difference: Lark Minutes handles mixed Chinese-English sentences naturally, for example “we need to review the QPS for this API”. It did not phonetically transliterate English words into Chinese characters, thanks to training data optimized for internet workplace scenarios.

Major downside: usability drops sharply outside the Lark ecosystem. Teams not using Lark must consider the workflow migration cost solely for transcription. Free tier storage is capped, so heavy users need to monitor quotas.

Rating: ★★★★☆ Scores Accuracy: 8.5/10 Feature completeness: 8/10 Scenario fit: 7.5/10 (strong ecosystem dependency) Value for money: 8/10

Tool 3: Tongyi Tingwu

One-sentence summary A multilingual transcription tool built on the Tongyi Qwen model family, with distinctive innovations in summary generation.

Core features

Real-time and multilingual transcription: Supports Chinese, English, Japanese, Cantonese and more.

Speaker sentiment tagging: Labels remarks as positive, questioning or neutral after identifying speakers.

Mind map generation: Breaks meeting content into mind-map structure.

Chapter preview and action item extraction: Automatically outlines discussion threads and pulls out action items.

Best use cases

Multilingual meetings, personal users needing fast content overview, negotiation reviews where speaker sentiment analysis helps.

Test experience

Tongyi Tingwu’s Mandarin transcription accuracy was comparable to the previous two tools. Its differentiation lies in post-transcription information processing.

Sentiment tagging judges tone based on filler words. For instance, “I don’t think this will work” was tagged as questioning. It offers some reference value but has false positives: ordinary questions were marked neutral, while hesitant statements were labelled questioning. Results should be treated as hints, not definitive conclusions.

Mind map outputs resemble a table of contents. They work well for logically structured project reviews but add limited value for loosely structured discussions. The daily free quota makes it friendly for infrequent users.

Rating: ★★★★☆ Scores Accuracy: 8.5/10 Feature completeness: 8.5/10 Scenario fit: 8/10 Value for money: 8.5/10

Tool 4: IFlytek Tingjian

One-sentence summary A mature transcription tool built on iFlytek’s speech technology, with long-term optimizations for multi-dialect support and custom professional term libraries.

Core features

High-precision transcription: High claimed accuracy for general Chinese scenarios. Custom term libraries boost recognition of niche industry terminology.

Dialect recognition: Supports many Chinese dialects plus over 200 languages.

Text polishing: Converts spontaneous spoken language into formal written style.

Best use cases

Interview transcription, meetings with accents or dialects, professional scenarios requiring formally formatted transcripts.

Test experience

IFlytek Tingjian delivered stable transcription. Preloaded custom term libraries (with product abbreviations imported in advance) recognized English terms well in the test recording.

Its text polishing function deserves special attention. It does not merely remove filler words. It rewrites conversational sentences such as “We actually discussed this feature before, but we didn’t make a decision back then” into formal prose: “This feature was previously discussed but no resolution was reached.” This saves revision time for meeting minutes ready for submission. However, polished transcripts lose some original tone, so they suit formal archiving rather than cases where nuance of speech must be preserved.

Full dialect capability was not thoroughly tested in this round, though the feature list shows broad coverage.

Rating: ★★★★☆ Scores Accuracy: 9/10 Feature completeness: 8.5/10 Scenario fit: 8/10 Value for money: 8/10

Comparative analysis

Under clear audio and standard Mandarin conditions, all four tools achieve similar transcription accuracy. Differences emerge in edge cases: IFlytek Tingjian has the most mature support for dialects and custom industry glossaries; Lark Minutes and Tongyi Tingwu handle mixed Chinese and English most naturally; Meetingminutes performs reliably for far-field microphone capture.

Ecosystem integration defines product positioning. Lark Minutes is tightly coupled with Lark Docs, Calendar and Tasks for teams within that platform. Tongyi Tingwu connects with Alibaba Cloud and offers API access. IFlytek Tingjian functions primarily as a standalone professional transcription tool. Meetingminutes prioritizes mobile and offline meeting scenarios.

Output formats shape downstream workflows. Tongyi Tingwu’s mind maps and sentiment tags serve reporting and review use cases. IFlytek Tingjian’s text polishing is ideal for formal documents. Lark Minutes sends transcripts straight into collaborative documents and eliminates file transfers.

On pricing, Tongyi Tingwu provides a daily free allowance, Lark Minutes includes basic free storage, while the others follow different pricing models. Selection depends on usage volume and required features.

Overall, tool selection comes down to matching your scenario. Alignment across four factors — ecosystem, language mix, output format requirements and usage frequency — matters more than accuracy scores alone.