Comparison
Sokuji alternative: open-source vs hosted
Two-way desktop speech translation — fair comparison.
Sokuji is the closest open-source neighbour to a paid desktop simultaneous interpreter. Parley is a commercial hosted desktop app. Neither replaces the other for every user.
In short: Sokuji is open-source real-time two-way speech translation (desktop on Windows, macOS and Linux, plus a Chrome/Edge extension). It can run with cloud providers when you bring your own API key, or fully offline on-device; translated audio goes into any app microphone, including Zoom, Meet and Teams. Parley is a commercial desktop app: hosted Qwen and Doubao, prepaid per session-minute, and the other people need not install anything. If you want to bring your own key, run local models and accept AGPL-3.0, Sokuji is a strong choice. If you want zero-config hosted translation and a per-minute wallet, Parley fits. Neither product is a unique option. (Disclosure: we make Parley.)
At a glance
| Dimension | Sokuji | Parley |
|---|---|---|
| Shape | Open-source desktop app + browser extension; virtual microphone into meeting apps and other software | Commercial desktop app for Windows and Mac; sits between you and mainstream meeting software |
| Open source & licence | Yes — AGPL-3.0; about 1398 GitHub stars as of 2026-10-15 | Proprietary commercial product |
| Billing | Software free; cloud usage billed by the provider you choose (or optional Kizuna AI metered account). Fully offline local inference needs no API spend | Prepaid wallet; Qwen $0.070/minute, Doubao $0.148/minute; $1 first top-up trial; no subscription. One connected session covers both directions |
| Bring-your-own API key | Yes — OpenAI, Gemini, Palabra.ai, Doubao AST 2.0, Soniox, OpenAI-compatible endpoints and others (README 2026-10-15). Optional Kizuna AI account can hold a key for you | Managed Qwen and Doubao out of the box (no self API key required). Optional personal API licence for your own keys; OpenAI and Gemini are Coming soon on managed list |
| Offline / on-device | Yes — first-class local inference (ASR → translation → TTS on device; no key required for that path) | No — managed translation needs internet and sign-in |
| Platforms | Windows, macOS, Linux desktop; Chrome and Edge extension | Windows and Mac desktop installers |
| Languages (documented) | README (2026-10-15): 99+ speech recognition, 55+ translation, 53 text-to-speech; site also states 35+ translation languages — confirm current figures on GitHub/site | Managed models Qwen and Doubao (live); see languages |
| Other participants | Need not install Sokuji; they hear ordinary meeting audio via your virtual microphone | Need not install Parley; they stay in their usual meeting app |
| Venue / in-person | Site documents face-to-face modes (one computer, earbuds or shared captions) | Separate venue mode for in-person talks (separate venue licence) |
| Ear return | Passthrough / monitor your own voice is documented; mix percentage is not a single public default figure here | Ear return on by default at about a 15% mix (adjustable); see FAQ |
Sources: Sokuji GitHub repository (README, AGPL-3.0 licence, ~1398 stars on 2026-10-15); Sokuji product site; Parley’s current pricing. Sokuji and related marks belong to their owners; no affiliation is implied.
When Sokuji is the better fit
- You want free software you can inspect and build. AGPL-3.0 source on GitHub; you can clone, modify and ship your own build under that licence.
- You need offline or on-device inference. Local ASR, translation and TTS without sending audio to a cloud provider — a first-class path, not only a fallback.
- You prefer bring-your-own keys or your own endpoint. Multiple cloud providers plus OpenAI-compatible endpoints; you pay the provider directly if you skip the optional Kizuna AI account.
- You need Linux or a browser extension. Desktop covers Linux; the Chrome/Edge extension targets supported meeting sites without a system virtual mic.
When a hosted tool like Parley fits
- You want zero-config hosted models. Download, top up a prepaid wallet, and use managed Qwen or Doubao without wiring your own API keys first.
- You want transparent per-minute pricing you control. Published rates, no subscription; stop the session when you are done. Silence while connected is billed.
- The other people should install nothing. Same idea as Sokuji’s virtual mic: counterparts stay in their meeting app. Parley packages that as a commercial Windows/Mac workflow with ear return (~15% by default).
- You want a separate venue / in-person product path. Parley documents venue interpretation with its own licence; useful when the purchase is “hosted minutes + venue”, not “self-host models”.
Related pages: native meeting translation vs Parley (Zoom / Teams / Meet overview); JotMe monthly captions vs desktop; translation earbuds vs software; eight tools compared; how to choose a meeting translation tool (includes a Sokuji section).
How Parley billing works
Managed translation needs wallet credit. One connected session covers both directions — speaking and listening share a single charge — and billing includes silence while the session is connected, so stop the session when you are done. Qwen is $0.070/minute and Doubao is $0.148/minute; a $1 first top-up trial is available. Only Qwen and Doubao are live today; OpenAI and Gemini are Coming soon. Ear return is on by default at about a 15% mix and can be turned down or off. Venue interpretation needs a separate venue licence. Windows installers are currently unsigned, so Windows may warn about an unknown publisher. See pricing and FAQ.
Setup
This page does not include click-by-click setup for Sokuji or Parley. For Sokuji, follow the project README and sokuji.kizuna.ai. For Parley, the app appears as a microphone and speaker path for mainstream meeting software on Windows and Mac — exact menus differ by OS and client version, so follow the download and audio checklist and test both directions before a real call. More on the workflow: meeting translation.
Questions
Is Sokuji a good Parley alternative (and the other way around)?
Yes, for many overlapping bilingual-meeting jobs — both can put translated speech into a meeting microphone so the other people need not install anything. They differ on licence (AGPL vs commercial), offline capability, whether you must bring your own API key, platforms (Sokuji includes Linux and a browser extension), and billing (provider/self-host vs Parley’s prepaid per-minute wallet). Choose by those priorities; neither product is a unique desktop two-way option.
Do I need to bring my own API key with Sokuji?
Not for the local on-device path: Sokuji’s README documents fully offline inference without an API key. For cloud models you bring your own key for the providers listed in the Sokuji README (checked 2026-10-15), or you can use the optional Kizuna AI account that holds a key and meters usage. Parley’s managed Qwen and Doubao path does not require your own key; OpenAI and Gemini are coming soon on Parley’s managed list, and an optional personal API licence exists if you prefer your own keys.
Can Sokuji run fully offline?
Yes. The repository and product site present on-device ASR, translation and text-to-speech as a first-class path — no key and no internet for that mode. Quality and resource use depend on the local models you download. Parley’s managed translation needs a network connection and sign-in.
Do the other participants need to install anything?
With either tool in the usual virtual-microphone meeting setup, no. You run the desktop app (or Sokuji’s extension on supported sites); the other people stay in their meeting app and hear ordinary call audio. Confirm current Sokuji extension site support on the Sokuji site.
When should I pick Sokuji over Parley?
Pick Sokuji when open source, AGPL, offline inference, Linux, a browser extension, or bring-your-own keys matter most. Pick Parley when you want hosted Qwen/Doubao without configuring providers first, a published per-minute wallet, and a Windows/Mac commercial installer with ear return at about 15% by default. Many people evaluate both; that is expected.
Try it on your next call
Download Parley for a short prepaid test — or try Sokuji from GitHub if open source or offline is your priority.
Download and set up → See pricing → JotMe compare → Native meeting translation compare → Sokuji on GitHub →