Multilingual & Mixed-Language Proceedings: Transcription and Translation
Transcribe depositions, interpreted hearings, and mixed-language recordings in 22 languages — with clearly labeled English machine translation alongside the preserved original.
The Multilingual Record Problem
Proceedings involving languages other than English are growing while the professionals who handle them are getting scarcer. NPR reported in 2025 that rising interpreter demand is forcing judges to delay hearings nationwide. California logged over 377,000 Spanish-interpreted court events in a single fiscal year against roughly 1,310 available Spanish interpreters, and has committed $6.8 million over five years to rebuild its interpreter pipeline. Meanwhile the US court reporter workforce shrank about 7% between 2019 and 2023.
For attorneys, the practical consequence is a growing pile of recordings — Spanish depositions, interpreted hearings, bilingual client interviews, wiretap audio — that are slow and expensive to turn into usable transcripts through traditional channels.
Why Generic AI Transcription Fails on Multilingual Audio
General-purpose transcription tools have a documented failure mode on multilingual recordings: covert translation. Many speech models silently translate non-English speech into English text — with no marker that translation occurred, no preserved original, and speaker attribution scrambled at the language switch. The transcript reads as if the witness spoke English. There is no way to tell which words were actually said, in which language, or how faithful the hidden translation was.
This is precisely what court-system guidance warns against. The National Center for State Courts' machine translation guide tells courts to disclose whenever machine translation is used and to treat its output as a starting point for human review — never a silent substitute for the original. We wrote about the covert-translation failure mode and how we engineered against it on the blog.
The American Bar Association points the same direction, in terms aimed directly at lawyers. The ABA Standards for Language Access in Courts treat qualified interpretation and translation as the foundation of meaningful access to the courts. ABA Formal Opinion 500 makes language access part of a lawyer's duties of competence and communication. And ABA Formal Opinion 512, the ABA's guidance on generative AI, requires lawyers who use AI tools to understand their limitations and supervise their output. A transcript that hides its own translation makes all three obligations impossible to meet.
One Recording, Two Languages
Interpreted proceedings change language constantly: question in English, answer in Spanish, rendering back into English. Bilingual client interviews drift mid-sentence. Jail calls switch depending on who is on the line. Most transcription tools force a single language across the whole file and mistranscribe everything that does not match it.
MatterScribe detects language changes inside a single recording and transcribes each stretch in the language actually spoken. There is no pre-splitting the audio, no separate upload for each language, and no one language forced across the file. Every second of audio belongs to exactly one run, so nothing is dropped at the seams between languages, and every segment keeps its own language, speaker label, and timestamp.
In mixed-language validation, one 73-minute interpreter-dense recording was divided into 49 separate language runs, each transcribed in the language actually used in that stretch.
Sustained passages in each language are what the system separates reliably. A stretch shorter than roughly six seconds is merged into the surrounding language rather than transcribed on its own, because transcribing eight seconds of Spanish with an English model and aligner produces worse output than absorbing a brief switch.
How MatterScribe Handles Multilingual Recordings
Original-language transcription in 22 languages. Tell MatterScribe which languages appear in the recording (or let auto-detect find them) and each segment is transcribed in the language actually spoken — Spanish testimony stays Spanish, with speaker labels and timestamps intact. The original is the record; it is never overwritten.
Labeled English machine translation. Enable translation and each non-English segment gets an English rendering displayed as a separate, clearly labeled machine-translation line beneath the original — on screen and in every export. A visible notice states that translations are machine translations, not certified translations. Nobody who reads the transcript can mistake translated text for spoken testimony.
Side-by-side review with word-synced playback. The Review Dashboard shows Original, Original + English, or English-focused views. Playback highlights the original-language text word by word — including character-level sync for Chinese and Japanese and right-to-left rendering for Arabic — so a bilingual reviewer can verify any line against the audio in seconds.
English protection by default. English-only recordings run with a cross-check that flags non-English speech rather than silently mistranscribing it. Your English recordings don't get spuriously "translated," and foreign-language audio gets surfaced instead of buried.
Foreign-Language Depositions, Jail Calls, and Other Recorded Evidence
Foreign language deposition transcription. When a witness is deposed in Spanish, Mandarin, or any other supported language, upload the recording and get the testimony transcribed in the language spoken, with a labeled English machine translation line under each answer. Review the same day instead of waiting weeks for a translation vendor.
Interpreted hearing and deposition transcripts. Recordings of interpreter-assisted proceedings switch languages on nearly every turn. MatterScribe keeps each turn in its original language with correct speaker attribution, so you can compare the witness's words against the interpreter's rendering.
Jail calls and intercepted recordings. Criminal defense teams routinely receive hours of foreign-language jail calls and intercepted communications, often accompanied by a transcript prepared by the other side. An independent, searchable original-language transcript with labeled machine translation gives you a fast first-pass review of what was actually said and helps you decide which recordings deserve a certified human translation.
Recorded statements and client interviews. Witness statements, investigator interviews, and bilingual client intake recordings get the same treatment: original language preserved, English translation clearly labeled, everything searchable.
Measured Accuracy, Not Marketing Claims
Before shipping multilingual support we ran a benchmark program against public corpora with human reference translations. The non-English cross-check caught 131 of 132 non-English test files (99.2% sensitivity) while flagging zero of 38 English files, including degraded appellate argument audio. Translation adequacy averaged chrF 64.1 across the supported languages (chrF is a standard machine-translation quality metric), reaching 75.0 for Spanish at full hearing length, with a Chinese character error rate of 9.1% on a 61-minute recording. DOCX export parity was verified line-for-line (105/105 translation lines) on a staged court-format run.
Certified Translation Still Has Its Place
MatterScribe produces working transcripts and labeled machine translations for attorney review and case preparation. They are not certified transcripts or certified translations. For filings that require certification, engage a qualified human translator or interpreter — the preserved original-language transcript gives them an exact, timestamped source to work from, which is more than a silent English-only AI transcript could ever offer.
Supported Formats and Security
Upload any common format: standard audio (MP3, WAV, M4A, FLAC and more), video (MP4, MOV, AVI, MKV), virtual-meeting exports from Zoom, Teams, and Google Meet, and native court recording formats including .TRM (ForTheRecord™) with no conversion required. All files are encrypted with AES-256 in transit and at rest, processed in US-based SOC 2-compliant data centers, and never used to train AI models.
Get Started
Upload a multilingual recording and see the original-language transcript with labeled English translation in minutes. MatterScribe's 14-day free trial includes 120 minutes of transcription.
Start Your Free Trial Contact Us
Frequently Asked Questions
Can MatterScribe transcribe a Spanish deposition or hearing recording?
Yes. MatterScribe transcribes recordings in Spanish and 21 other languages, preserving the original-language text segment by segment with speaker labels and timestamps. You can also add a clearly labeled English machine translation alongside the original — the Spanish testimony itself is never overwritten.
Can I translate a court hearing recording into English?
Yes. Upload the recording, declare the languages spoken, and enable English translation. Each non-English segment gets an English translation shown as a separate, clearly labeled machine-translation line beneath the original text, on screen and in exports. It is machine translation — not a certified translation.
What happens when a recording switches languages, like an interpreted proceeding?
MatterScribe handles mixed-language recordings automatically: it transcribes each portion in the language actually spoken, including recordings that alternate between English and another language throughout, such as interpreter-assisted testimony. Every segment keeps its original language, speaker label, and timestamps.
Is the translation a certified translation?
No. Translations are machine translations, clearly labeled as such, and are not certified translations. They are designed for attorney review and case preparation. For filings that require a certified translation, engage a qualified human translator — the preserved original-language transcript gives them an exact source to work from.
Can I use MatterScribe for jail calls or other foreign-language recorded evidence?
Yes. Attorneys use MatterScribe to review foreign-language jail calls, intercepted communications, recorded statements, and client interviews. You get a searchable original-language transcript with clearly labeled English machine translation — an independent working view of what was said. Certified human translation remains the standard for transcripts offered into evidence.
Which languages does MatterScribe support?
22 languages: Spanish, Chinese (Mandarin), Korean, Russian, Arabic, Portuguese, French, German, Italian, Japanese, Polish, Dutch, Ukrainian, Greek, Turkish, Romanian, Czech, Hungarian, Finnish, Swedish, Norwegian, and Danish — plus automatic language detection and support for recordings that mix multiple languages.