# The best AI note takers for bilingual client meetings

By [David Wilson](https://testbenchlab.com/authors/david-wilson/). Updated 2026-11-02. US prices checked 2026-10-29.

If your client meetings run in two languages at once — English and Spanish in the US, English and French in Canada, English and Japanese or Mandarin with Asian clients — most AI note takers will let you down. The typical tool transcribes one language per meeting, so every sentence spoken in the second language comes out as nonsense. Notta is the note taker we would use for bilingual client meetings: you choose two of its 23 supported languages, and it transcribes both in live recordings, through the Notta Bot, and on uploaded files, with live translation alongside. Every plan includes 4 bilingual uses a month, and regular use needs the bilingual add-on at $9 a month (billed annually) on a paid plan. Fireflies.ai is the alternative for automatic language detection if you want automatic word-level language detection across 60+ languages, and Granola is the pick if you want bot-free notes with a desktop multi-language mode. Whatever you choose, test it with a recording of your own language pair first: code-switching is still hard for all speech recognition. If your meetings are each held in a single language rather than mixed, see our guide to the best AI note taker for multilingual sales calls.

**How we assessed:** These picks rest on vendor documentation rather than hands-on measurement. Before standardising on any tool, record 10 minutes of your real code-switching — a meeting that resembles an actual client call, including a few mid-sentence switches, since that is where errors cluster — and run the same file through every tool you are comparing. Mark each switch point in a reference transcript and count how many each tool got right, note the summary language each tool produced and whether you can change it, and if you are testing Notta, keep an eye on how many of your monthly bilingual uses the test consumes. Ten minutes of honest testing will tell you more about your language pair than any feature list.

## Quick list

| # | Award | Product | Score | Price |
|---|---|---|---|---|
| 1 | Best overall for bilingual client meetings | Notta |  |  |
| 2 | Best automatic language switching | Fireflies.ai |  |  |
| 3 | Best bot-free | Granola |  |  |
| 4 | Best budget two-language setup | Sembly AI |  |  |

## 1. Best overall for bilingual client meetings: Notta

Notta is the only pick that handles two chosen languages across every capture method you are likely to use with clients: real-time recording, the Notta Bot joining your meeting, and uploaded files (up to 1 GB or 3 hours of audio). You pick two of 23 supported languages — including Spanish, French, Japanese, Chinese (Simplified and Traditional), Korean, Portuguese, German, Italian, and more — and both are transcribed in the same meeting. Live translation sits alongside the transcript, so you can follow a client speaking the other language in real time (translation is not available on uploads). Transcripts can be translated into 42 languages, which helps when the client deliverable needs to be in one language and your internal notes in another. Every Notta plan includes 4 bilingual uses a month, so you can try the mode before committing; regular use needs the bilingual add-on at $9 a month billed annually ($108 a year) or $15 month-to-month, covering up to 200 bilingual uses a month on Pro and Business plans. Notta Pro itself is $8.17 a month, billed $97.99 a year.

## 2. Best automatic language switching: Fireflies.ai

Fireflies' Multi-Language Mode (beta) takes a different approach: instead of choosing two languages, it detects language word by word across 60+ languages, so you do not have to predict which pair your meeting will use. That makes it the strongest option for meetings where the language mix is unpredictable or involves more than two languages. The trade-offs: Multi-Language Mode is only available on Business and Enterprise plans ($19 per seat a month, billed annually, or $29 monthly), and the meeting summary comes out in the dominant language rather than a language you choose. Fireflies also advises not changing the meeting language after processing, so the setup matters. If your client meetings genuinely span many languages rather than a fixed pair, Fireflies is worth the premium.

## 3. Best bot-free: Granola

Granola is the pick for consultants who do not want a bot in the room. Its desktop multi-language setting transcribes all 31 desktop languages in one meeting, and the summary can be set to English or Auto (which follows the majority language). On mobile, Granola picks a single language per meeting, so the bilingual workflow is desktop-first. The Basic plan is free and the Business plan is $14 per user a month. If your clients are wary of meeting bots and your bilingual meetings happen at your desk, Granola covers the two-language case without an add-on — but if you also need to transcribe phone calls or uploaded recordings in two languages, Notta's broader capture methods make it the more complete choice.

## 4. Best budget two-language setup: Sembly AI

Sembly supports up to two languages per meeting or workspace, and mixed-language meetings work on every plan, including Basic at $10 a month billed annually ($17 monthly). Notes come out in the most-used language. That makes Sembly the cheapest way to get a genuine two-language setup with no add-on, though it lacks Notta's live translation and upload flexibility, and it does not offer Fireflies' word-level auto-detection. For a small team testing the bilingual workflow before investing more, Sembly Basic is a sensible starting point.

## Also considered

- [Microsoft Teams (Copilot / Teams Premium)](https://support.microsoft.com/topic/650cb6d2-8a33-40e7-840d-36bb90216aa4): Teams' multilingual speech recognition lets each participant set their own spoken language across 9 listed languages, but translated transcripts are only available during the meeting and are not kept afterwards. Requires a Copilot licence or Teams Premium; pricing unverified.
- [Sonix](https://sonix.ai/pricing): Offers 'Transcribe in Multiple Languages' for uploaded files across 54+ languages, but it is a file-based workflow rather than a live meeting note taker. Core plan $25 a month or $275 a year (5 hours a month), extra hours $10 each.
- [tl;dv](https://tldv.io/pricing/): Uses a Whisper model on Business and Enterprise plans (€29 per seat a month, billed annually) and its help material suggests support for multiple languages in one meeting, but we could not fully verify the details before publication.
- [Otter.ai](https://otter.ai/pricing): Supports 6 languages with one transcription language per session. A French-English simultaneous mode has been reported but is unverified, and Otter's own launch guidance said speakers must use the language being transcribed. Pro is $8.33 a month billed annually.

## How we chose

We looked for tools with a mixed-language mode documented by the vendor — not just a long list of supported languages, since most tools with 100+ languages still transcribe only one language per meeting. Language count alone was not a criterion. We then checked whether each mode works on live meetings, meeting bots, and uploaded files; whether the summary language can be controlled; and what the mode costs on the plan that includes it. Prices and plan details were checked on 29 October 2026. These recommendations are based on vendor documentation of each tool's mixed-language mode rather than hands-on measurement of code-switched recordings, so test your own language pair before standardising on a tool.

## Which AI note taker handles two languages in one meeting?

Notta is the note taker we would use for bilingual client meetings: you choose two of 23 languages and it transcribes both, whether you are recording live, running the Notta Bot, or uploading a file. Only a handful of tools have a true mixed-language mode rather than a long single-language list. Here is how the main options compare.

| Tool | Mode | Languages in mode | Live / bot / upload | Summary language | Plan and price |
|---|---|---|---|---|---|
| Notta | Choose two languages | 23 | Live, bot and upload | Transcripts translatable into 42 languages | 4 free bilingual uses/mo on every plan; add-on $9/mo annual on Pro ($8.17/mo billed $97.99/yr) or Business |
| Fireflies.ai | Auto-detect (word-level) | 60+ | Live and bot | Dominant language | Business $19/seat/mo annual ($29 monthly) |
| Granola | Multi-language (desktop) | 31 desktop; mobile one per meeting | Live, bot-free desktop | English or Auto (majority) | Basic $0; Business $14/user/mo |
| Sembly AI | Choose two languages | 2 per meeting or workspace | Live and bot | Most-used language | Basic $10/mo annual ($17 monthly) |

## What is the difference between bilingual mode and auto-detect?

Auto-detect usually picks one dominant language for the whole meeting; a bilingual or multi-language mode transcribes both. The distinction matters more than any language count on a pricing page. In a bilingual mode like Notta's, you tell the tool which two languages to expect, and it transcribes both — predictable, but it assumes you know your pair in advance. In an auto-detect mode like Fireflies' Multi-Language Mode, the tool decides language word by word, which handles unpredictable mixes but puts the burden on the detection model. Fireflies' own help documentation distinguishes the two approaches. For client work with a known pair — say English and Spanish — a chosen bilingual mode is the more reliable starting point; for meetings where you cannot predict the mix, auto-detect earns its premium.

## Can it handle people switching language mid-sentence?

Fireflies documents word-level language detection, which is designed for exactly this; the other vendors do not say how their modes behave at the switch point. The honest answer is that mid-sentence code-switching is where every tool is weakest. Research bears this out: the SwitchLingua dataset (NeurIPS 2025) showed state-of-the-art speech recognition still has large error gaps on code-switched speech, and the PIER metric (ICASSP 2025) found that standard word error rates can improve while the code-switched words themselves get worse. In practice, that means the transcript may look accurate overall while the exact words your client switched languages to say — often the emotionally important ones — are the ones that get mangled. This is why we recommend testing any tool with your own recording before standardising on it, and why speaker identification and overlapping speech matter too: see our guides on why speaker identification matters and why transcription struggles with overlapping speech.

## What language will the summary be in?

In most tools, the summary comes out in the dominant language of the meeting. Fireflies summarises in the dominant language; Sembly writes notes in the most-used language; Granola lets you choose English or Auto (majority). Notta's transcripts can be translated into 42 languages, which gives you a way to produce the deliverable in the client's language even when the meeting ran mostly in yours. Our practical advice for client work: fix the summary language before the meeting, and check what the tool actually produced before anything goes to the client. A summary that flips language halfway through a project is a small thing that reads as carelessness.

## Does Microsoft Teams or Zoom do this already?

Teams' multilingual speech recognition lets each participant set their own spoken language — 9 languages are listed, including English, Spanish, French, German, Italian, Portuguese, Chinese, Japanese, and Korean — with a Copilot licence or Teams Premium. The catch: translated transcripts are only available during the meeting and are not kept afterwards, so you cannot rely on them as a record. If your clients are all on Teams with the right licences, it is worth trying before buying another tool. But if you need a persistent bilingual transcript, a summary you can edit before sharing, or support for calls outside Teams, a dedicated note taker like Notta still earns its place.

## What does bilingual transcription cost?

Prices checked 29 October 2026. The cheapest route to a genuine two-language setup is Sembly Basic; the most complete capture is Notta with its add-on; the auto-detect route is Fireflies Business.

| Setup | Plan | Price (USD, annual billing) |
|---|---|---|
| Notta Pro + bilingual add-on | Pro + add-on | Pro $8.17/mo (billed $97.99/yr) + add-on $9/mo ($108/yr); 4 free bilingual uses/mo on every plan |
| Fireflies Business | Business | $19/seat/mo ($29 monthly); Multi-Language Mode is Business and Enterprise only |
| Granola Business | Business | $14/user/mo; Basic is free |
| Sembly Basic | Basic | $10/mo ($17 monthly) |

## How do I test a note taker with my languages?

Record 10 minutes of your real code-switching and check every switch point. Use a meeting that resembles an actual client call — not a clean read of single-language sentences — and include at least a few mid-sentence switches, since that is where errors cluster. Run the same recording through every tool you are comparing, mark each switch point in a reference transcript, and count how many each tool got right. Note the summary language each tool produced and whether you can change it. If you are testing Notta, keep an eye on how many of your monthly bilingual uses the test consumes. Ten minutes of honest testing will tell you more about your language pair than any feature list.

## Frequently asked questions

### Can AI transcribe Spanglish?

Partially. Tools with a true bilingual mode — Notta with two chosen languages, or Fireflies with word-level auto-detect — will transcribe the English and Spanish portions of Spanglish, but mid-sentence switches are where speech recognition is weakest, and research shows errors cluster exactly on the code-switched words. Test with your own recording before relying on it.

### Does Otter.ai transcribe two languages at once?

Not reliably. Otter supports 6 languages with one transcription language per session. A French-English simultaneous mode has been reported but is unverified, and Otter's own guidance has said speakers must use the language being transcribed. For genuine two-language meetings, Notta, Fireflies, Granola, and Sembly are stronger choices.

### Can an AI note taker translate a bilingual meeting live?

Notta offers live translation alongside its bilingual transcription for live recordings and meetings, so you can follow a client speaking the other language in real time — though translation is not available on uploaded files. Most other note takers translate the transcript after the fact rather than live. For live translated captions aimed at audiences rather than note taking, see our guide to live caption translation.

### Which languages does Notta's bilingual transcription support?

Notta's bilingual mode covers 23 languages: Arabic, Chinese (Simplified and Traditional), Dutch, English, French, German, Hindi, Indonesian, Italian, Japanese, Javanese, Korean, Nepali, Polish, Portuguese, Romanian, Russian, Spanish, Thai, Turkish, Ukrainian, and Vietnamese. You choose two per meeting. Note that some of Notta's own marketing pages still say '10+ languages', but the current documentation lists 23.

### Why does my transcript turn one language into gibberish?

Because the tool was transcribing in a single language. Most AI note takers pick one transcription language per meeting, so anything spoken in the second language is forced through the wrong model and comes out as nonsense. The fix is a tool with a true bilingual or multi-language mode — Notta, Fireflies' Multi-Language Mode, Granola's desktop multi-language setting, or Sembly's two-language setup.

## Related reading

- [The best AI note taker for multilingual sales calls](https://testbenchlab.com/best/best-ai-note-taker-for-multilingual-sales-calls/)
- [The best AI note takers for client meetings in 2026](https://testbenchlab.com/best/best-ai-note-taker-for-client-meetings/)
- [Live translated captions for lectures: apps that work today](https://testbenchlab.com/best/live-translated-captions-for-lectures/)
- [Why speaker identification matters in interview transcripts](https://testbenchlab.com/guides/why-speaker-identification-matters/)
- [Why transcription tools struggle with overlapping speakers](https://testbenchlab.com/guides/why-transcription-struggles-with-overlapping-speech/)
- [Notta Memo review](https://testbenchlab.com/reviews/notta-memo-review/)

## Sources

1. [SwitchLingua: the first large-scale multilingual and multi-ethnic code-switching dataset](https://neurips.cc/virtual/2025/poster/121687) — NeurIPS 2025 (Xie et al.); Research; published 2025
2. [PIER: a novel metric for evaluating what matters in code-switching](https://arxiv.org/abs/2501.09512v2) — ICASSP 2025 (Ugan et al.); Research; published January 2025
3. [A benchmark for in-the-wild African code-switched speech recognition](https://arxiv.org/pdf/2608.26434) — arXiv (AfriSwitch); Research; published 2026
4. [Bilingual transcription support article](https://support.notta.ai/hc/en-us/articles/23437010425755) — Notta; Vendor documentation; checked 29 October 2026
5. [Transcribe Fireflies meetings in multiple languages with Multi-Language Mode (beta)](https://guide.fireflies.ai/articles/2585231364-transcribe-fireflies-meetings-in-multiple-languages-with-multi-language-mode-beta) — Fireflies.ai; Vendor documentation; checked 29 October 2026
6. [Multi-language settings](https://docs.granola.ai/help-center/customising-granola/multi-language) — Granola; Vendor documentation; checked 29 October 2026
7. [Choose languages for your Sembly meetings](https://help.sembly.ai/en/articles/124-choose-languages-for-your-sembly-meetings) — Sembly AI; Vendor documentation; checked 29 October 2026
8. [Multilingual speech recognition in Teams](https://support.microsoft.com/topic/650cb6d2-8a33-40e7-840d-36bb90216aa4) — Microsoft; Vendor documentation; checked 29 October 2026