Why Language Matters
Speech recognition accuracy depends heavily on knowing which language is being spoken. When 12scribe's transcription engine knows the correct language in advance, it uses the right phonetic model, vocabulary, and grammar rules to interpret the audio. The result is a dramatic difference in quality: a correctly specified language typically produces 94% or higher word accuracy, while an incorrectly matched language can drop accuracy to 80% or below — sometimes producing entirely garbled text.
This happens because speech sounds overlap across languages. The English word "yes" and the Russian word "ес" sound nearly identical, but they mean different things and follow different grammatical patterns. When the engine knows it is processing English, it makes the right choice. When it guesses wrong, errors cascade through entire sentences.
Setting the language takes only a few seconds, but the impact on transcription quality is substantial. It is one of the simplest things you can do to get significantly better results from 12scribe's transcription system.
Per-Project vs Global Setting
12scribe gives you two levels of control over the transcription language:
- Global default — Set your preferred language in the app's main settings. Every new project you create will inherit this language automatically. If you primarily record in one language, set the global default once and forget about it.
- Per-project override — Each project can have its own language setting that overrides the global default. This is useful when you work in multiple languages — for example, an English-language business meeting project alongside a Russian-language lecture project.
To set the global default, go to Settings and find the Transcription Language option. To change the language for a specific project, open the project, tap Project Settings, and select the language there. The per-project setting takes priority whenever it is set. See Project Settings for the full list of per-project options.
If you forget to set the language before recording, you can still change it before transcription starts. The language setting is applied at transcription time, not at recording time. So even if you recorded first and realize the language is wrong, open project settings, correct the language, and then start transcription — the result will use the correct language model.
Supported Languages
12scribe currently supports transcription in the following languages:
- English — optimized for American, British, Australian, and other major English accents
- Russian — full support for standard Russian speech
Each language model is trained on large datasets of real-world speech, covering a range of accents, speaking speeds, and audio conditions. The models handle domain-specific vocabulary reasonably well — technical terms, proper nouns, and jargon are usually transcribed correctly if they appear in the training data. Uncommon terms may occasionally be misrecognized, but the surrounding context typically makes the intended word clear.
Language support is expanding. As new language models become available, they are added to both the global settings and the per-project language picker. Check the app's settings periodically to see if your language has been added.
The transcription engine runs on 12scribe's backend servers, which means language models do not consume space on your device. Whether you use one language or all supported languages, there is no local storage overhead. Learn more about how the server-side process works in How Transcription Works.
Mixed-Language Recordings
Real-world recordings often contain more than one language. A professor might switch between English and Russian, or a meeting participant might quote a source in a different language. How does 12scribe handle this?
Currently, each transcription run uses a single language model. If you set the language to English and the speaker switches to Russian mid-sentence, the Russian portions will be transcribed with the English model, resulting in poor accuracy for those segments. The English portions will still be accurate.
For recordings with significant language mixing, here are practical strategies:
- Choose the dominant language — Set the language to whichever language makes up the majority of the recording. The primary content will be transcribed accurately, and short switches to the other language will appear as approximate transliterations.
- Use markers to flag switches — When you hear the speaker switch languages during recording, add a marker. After transcription, you can manually correct those marked segments rather than searching through the entire transcript.
- Split into separate projects — If a recording has long, distinct sections in different languages (for example, a 30-minute English presentation followed by 30-minute Russian Q&A), consider using auto-transcription on separate recordings, each with the correct language setting.