Juristiem un tiesām

MultiVoice.AI ir izstrādājis īpašu moduli tiesas sēžu protokolu atšifrēšanai. Sistēma ir pārbaudīta tiesas sēžu protokolu atpazīšanai Latvijas tiesās (civillietās, krimināllietās, administratīvajās lietās) (latviešu valodā) un Krievijas šķīrējtiesās (krievu valodā). Publiski pieejams 2025. gada decembrī.

Mūsu mākslīgais intelekts apstrādā failus. Apstrādes laiks ir atkarīgs no dažādiem parametriem. Ja nepieciešams (piemēram, neskaidras runas gadījumā), mākslīgais intelekts transkripcijas procesam pieslēdz personu (mūsu darbinieku, dzimtās valodas runātāju).

2025. gada 18. augustā Intelektuālā īpašuma tiesas (Krievija) prezidijā notika lietas C01-2093/2023 izskatīšanas sēde. Viens no lietas dalībniekiem, izmantojot mūsu pakalpojumu, pasūtīja tiesas sēdes audioieraksta transkriptu.

Processing challenges:

1) Some participants were present in person at the court hearing, while others connected via video conference. 2) During the first part of the court hearing, the judges’ microphone volume was low, so the audio recording of the hearing is poorly audible, while the second half is clearly audible. As a result, the audio file has different volume levels. 3) Two judges and four representatives spoke at the court hearing—a total of six different people. We needed to identify the six speakers. The hearing lasted 32 minutes.

Result:

Our service created a 100% accurate transcript of the audio recording within 6 hours and 20 minutes of the order. The cost to the client was 61.35 euros.

1) Daži dalībnieki tiesas sēdē piedalījās klātienē, bet citi sazinājās, izmantojot videokonferenci. 2) Tiesas sēdes pirmajā daļā tiesnešu mikrofonu skaļums bija zems, tāpēc tiesas sēdes pirmā puse audioierakstā ir grūti dzirdama, savukārt tiesas sēdes otrā puse ir viegli dzirdama. Tā rezultātā audiofailā bija dažādi skaļuma līmeņi. 3) Tiesas sēdes laikā uzstājās 2 tiesneši un 4 pārstāvji. Kopā 6 dažādi cilvēki. Bija jānosaka 6 runātāji. Sēdes ilgums bija 32 minūtes.

Processing challenges:

1) Some participants were present in person at the court hearing, while others connected via video conference. 2) A Russian-speaking witness was present at the court hearing. Therefore, a Russian-Latvian interpreter was present. The court hearing was effectively conducted in two languages: Latvian, and then the interpreter translated everything into Russian. 3) The interpreter translated via video conference, and the interpreter’s audio level was twice as loud as the sound in the courtroom. As a result, the audio file had different volume levels. 4) One judge, an interpreter, and four other people spoke at the court hearing. That’s six different people in total. It was necessary to identify the six speakers. The meeting lasted 1 hour and 12 minutes.

Result:

Our service created a 100% accurate transcript of the audio log within 17 hours and 15 minutes of the order. The AI ​​had to involve a human (a professional native speaker of Russian and Latvian) for the transcription, as the AI ​​was unable to simultaneously recognize speech in two different languages ​​at different volume levels.

The cost to the client was 9,348 Russian rubles or 98.40 euros.

Tiesas sēde Rīgas apgabaltiesā notika 2025. gada 3. jūnijā. Viens no lietas dalībniekiem, izmantojot mūsu pakalpojumu, pasūtīja tiesas sēdes audioieraksta transkriptu. Latvijā tiesas sēdes notiek latviešu valodā.

If only one language is spoken in a courtroom (for example, only Latvian or only Russian), then speech recognition is very good. It also depends on the quality of the court’s audio recording. If the audio recording is barely audible, speech recognition will be poor.

Only you will have access to the files you upload. Once the files are processed, you can delete them from our system.