How to make SRT subtitles: local audio transcription in the browser
Subtitles for recordings are needed regularly: lectures, interviews, conference calls, rough podcasts. Manual marking takes time, and online services often require registration or send files to a third-party server. In this material - a practical scenario for working with the "Subtitles" tool, which recognizes speech directly in the browser and outputs a ready-made SRT with time marks. We will break down how to prepare the audio, what exactly to enter in the form, how to check the result and where the limitations lie. The scenario is fictional and presented as an example, not as a testing report.
Open tool: Subtitle generator ↗What exactly does the tool do and what is its feature
The "Subtitles" tool solves one specific task: convert an uploaded audio file into a text transcription with time marks and export it in SRT format. The key feature - local sound processing in the browser. This means that the audio is not sent to an external server for recognition: transcription is performed on your device. This approach reduces the risk of leaking confidential recordings and speeds up work with a stable connection, since you do not need to wait for the file to be loaded on a remote machine.
The interface is extremely simple: a web form with text fields, output language selection, a launch button and a separate panel with the result. No sliders, style presets, settings panels and comparison modes. The result can be copied or downloaded as plain text TXT, and subtitles - as SRT. This is a conscious limitation: the tool does not try to be a video editor, it covers the basic task - getting marked text.
It is important to understand that the recognition model may make mistakes: names, terms, numbers and strong accents often require editing. The queue and availability of the service may change, so for large volumes it is better to plan the work in advance. Below - a step-by-step scenario that helps get a predictable result even taking into account these limitations.
Example task: "Make subtitles for this recording" - but first you need to upload audio and choose a language.
Audio preparation: format, length and quality
Before uploading, check that the file meets the limits: browser audio - up to 20 MB and not longer than 5 minutes. If the recording is longer, it will have to be split into fragments and processed in parts, and then SRT stitched manually or with a script. For short interviews and notes this is usually enough, but for hour-long lectures you will need to cut it. Format better to choose common: WAV, MP3 or OGG. Compressed with a strong bitrate files are recognized worse.
Sound quality affects accuracy more than length. Remove background music, fan noise and echo if possible. Record on one microphone, not several remote ones. If there are multiple speakers in the recording, the tool does not separate them by voices: in SRT there will be a continuous text with timestamps. For dialogues this means that after transcription you will have to manually add names or marks. This limitation is worth considering in advance.
Another practical point: before uploading listen to the first 30 seconds and make sure the speech is understandable. If you do not understand the words at this stage, the model will also make a mistake. Sometimes helping with mild volume normalization, but do not overdo it: aggressive processing may add artifacts.
File interview.mp3, 4 minutes 20 seconds, 12 MB, one speaker, no music - suitable for upload.
Step-by-step workflow in the interface
Step 1. Open the "Subtitles" tool in the browser. Step 2. Upload an audio file, observing the limits: up to 20 MB and up to 5 minutes. Step 3. Select the output language - this is important because the dictionary and punctuation rules depend on the language. Step 4. If necessary, add context in the text field: names, terms, abbreviations. The field is limited to 8000 characters, but a few lines are enough for hints.
After completion you will see the transcription with time marks. It can be copied as plain text or downloaded as TXT. For subtitles select SRT export: this is the standard format understood by video editors, players and publishing platforms. SRT contains block number, start and end time, and the text of the replica. If you plan to embed subtitles in video, check that timestamps do not overlap and do not have negative values.
Do not try to use the tool as an editor: there are no settings panels, change highlighting and special comparison modes. All edits are made after export - in a text editor or in a subtitle program. This is a normal workflow: first rough transcription, then proofreading.
In the context field you can write: "Names: Anna, Dmitry. Terms: SRT, timestamp, transcription". This increases the chances of correct writing.
Result verification: what to look at first
First, check the timestamps. Open the SRT in a player and go through several random fragments. Make sure the subtitles appear synchronously with the speech and do not "fly" to the end of the recording. If the shift accumulates, perhaps the audio was cut or has variable sample rate. In this case try to re-encode the file to WAV with constant sample rate and repeat the recognition.
Second - text. Check proper names, numbers, dates and special terms. The model may confuse similar sounding words, especially in noisy recordings. Make a list of key words and search through it. If there are numbers spoken in words in the recording, verify them with the context: "fifteen" and "fifty" sound similar, and an error changes the meaning.
Third - splitting into replicas. SRT should not contain too long strings: it is inconvenient to read them. After export you can break long blocks manually or with a script. Also check that there are no empty blocks and duplicate timestamps. These simple checks take a few minutes, but significantly improve the quality of the final file.
Verification: open SRT in VLC, turn on subtitles and compare the first 30 seconds and the last 30 seconds. If the discrepancy is more than a second - re-encode the audio.
Tool limitations and how to work around them
Main limitations: audio up to 20 MB and up to 5 minutes, lack of batch processing, inability to automatically save and integrations. No export to DOCX or PDF, no OCR, no settings panels and comparison modes. Cannot publish the result directly or send by email. These are not flaws, but boundaries of the task: the tool does one thing - transcription to SRT. If you need another format, convert SRT after export.
For long recordings use cutting. Split the audio into fragments of 4-5 minutes, process each separately, then combine SRT. When combining do not forget to shift the timestamps of subsequent fragments by the duration of the previous ones. This can be done manually or with a simple script. Such an approach requires discipline, but allows working with lectures and long interviews.
Another limitation - accuracy. The model may make mistakes, and this is normal. Do not expect a perfect result without proofreading. If the recording has a strong accent, technical jargon or multiple speakers at the same time, accuracy decreases. In such cases context in the text field and subsequent manual editing helps. The tool does not replace an editor, it speeds up rough work.
For a 40-minute lecture: cut into 8 fragments of 5 minutes, process, then combine SRT with timestamp shift.
Practical scenario: from recording to ready SRT
Imagine a fictional scenario: you have a client meeting recording lasting 4 minutes 30 seconds. You want to get subtitles for an internal report. First check the file: size 14 MB, MP3 format, one speaker, light background noise. Upload audio to the tool, select language, add participants' names and project name to the context field. Start recognition.
After completion download SRT and open it in a text editor. Check timestamps and key terms. If you noticed errors, fix them manually. Then open SRT in a player and make sure subtitles are synchronized. If everything is okay, the file is ready for use. If the recording is longer than 5 minutes, cut it in advance into parts and repeat the process for each.
This scenario shows that the tool is well suited for short recordings and rough transcription. It does not require registration, does not send audio to the server and gives results in a convenient format. But it does not replace a full subtitle editor and does not guarantee perfect accuracy. Treat it as a quick assistant, not as a final solution.
Ready SRT: 1\n00:00:01,000 --> 00:00:04,200\nGood day, thank you for finding the time.\n\n2\n00:00:04,500 --> 00:00:08,100\nLet's discuss the project.
The "Subtitles" tool is a practical way to get a rough transcription with timestamps without registration and without sending audio to third-party servers. It is suitable for short recordings up to 5 minutes and 20 MB, works well with clean speech and gives results in SRT format. The main limitations - length, lack of batch processing and possible recognition errors. Use context in the text field, check timestamps and key terms, and for longer recordings, cut the audio into fragments.
Open tool: Subtitle generator ↗Interface translations are machine-assisted and may need refinement.