From audio to text
Transcription Service

Tips for Perfect Transcription

The magic is in the details

A high-quality transcription always follows a specific set of rules. These rules can often cause problems for artificial intelligences (AI), as they involve semantic subtleties in the language. While the initial transcription by AI is a true time-saver and speeds up a large portion of the work, it's important for a human to take control of an AI-generated transcription. This is where Medial Post-Editing by a professional translator comes into play. Only through this fine-tuning is it possible to achieve a high-quality transcription.

At tolingo, we have established the following transcription rules:

Basic Transcription Rules

1. The text is transcribed exactly as spoken. No corrections are made, meaning errors (e.g., grammatical errors in sentence structure) are included.
2. All statements, including seemingly unimportant filler words (e.g., "I say" or "so to speak"), are captured.
3. Unclear sections are marked as [unclear] in the continuous text.
4. Texts or sections in a foreign language are marked as [foreign language].
5. If there are statements from multiple unidentified speakers in an audio file, they are labeled as Speaker 1, Speaker 2, etc.
6. No timestamps are set for a simple transcription.

Advanced Transcription Rules

1. The text is transcribed exactly as spoken. No corrections are made, meaning errors (e.g., grammatical errors in sentence structure) are included.
2. All statements, including seemingly unimportant filler words (e.g., "I say" or "so to speak"), are captured.
3. Dialects are corrected (e.g., "haben wir" instead of "hamma").
4. All non-verbal sounds made by speakers (e.g., stuttering, "uhms," "ne?") are omitted.
5. Contentually irrelevant listener confirmations (e.g., "Hm," "Yes," "Oh, really") are also omitted. These are only transcribed in rare cases if they contribute to the meaning (e.g., as a response to a question).
6. Special events are placed in parentheses (e.g., (sound disturbance) or (phone ringing repeatedly)).
7. Only direct speech is placed in quotation marks (e.g., I asked him: "Why are you doing that?").
8. To avoid long sentences over several lines, punctuation is placed logically. A conjunction (e.g., "And") may appear at the beginning of a sentence.
9. Polite pronouns like "Sie" and "Ihnen" are capitalized. For example, when people in an interview address each other ("I have one more question for you."). When the interviewees use the informal "du," the "du" and all its forms (e.g., "dir," "dich," "dein") are written in lowercase.
10. The capitalization of foreign words follows the same rules as for their German equivalent, i.e., verbs in lowercase and nouns in uppercase (e.g., "I googled Cyberspace.").
11. All numbers from one to twelve are written out, and from thirteen onward, they are written as digits. Meaningful exceptions, such as dates, are also written as digits (e.g., "3.1.2017").
12. If there are statements from multiple unidentified speakers in an audio file, they are labeled as Speaker 1, Speaker 2, etc.
13. Incomplete sentences are marked with a "-" (e.g., "So, then they were-, no, again: There were four people in the-."). After the "-" regular punctuation is used. The "-" is placed directly after the word, without a space.
14. Incomplete words are only included if they provide meaningful content. Otherwise, they are considered stuttering and are omitted.
15. Pauses longer than four seconds are marked with the number of seconds in parentheses, e.g., for a seven-second pause: (7 sec.).
16. Words where the wording is unclear and only assumed are marked with a question mark and placed in parentheses (e.g., (?expert)). If the word seems nonsensical and logically does not fit, the section is marked as unclear (see the next point).
17. Timestamps are placed by the AI before the name/speaker and are checked by the Medial Post-Editor.

These rules form the foundation for flawless results and consistent quality of our transcriptions. The combination of AI and human expertise provides the best possible outcome in the shortest amount of time. This ensures that you receive a high-quality product at a great price.

→ Also check out our comprehensive practical guide: Avoid mistakes in interview transcription and our brief article on what transcription is.