ComIQ Audio & Video Transcriber

Fast speech-to-text for uploaded audio/video or public media URLs, with timestamps, SRT, VTT, QA-gated delivery, and hardware-accelerated local processing.

ComIQ Audio & Video Transcriber converts uploaded speech audio/video and supported public media URLs into structured transcripts through the ComIQ fulfillment engine. It is designed for meetings, interviews, podcasts, voice notes, presentations, lectures, customer calls, recorded videos, and automation workflows that need machine-readable transcripts.

The service returns the full transcript plus timestamped segments, SRT/VTT subtitles, language, processing metrics, execution-device metadata, and a ComIQ QA screening result.

Why use ComIQ Audio & Video Transcriber?

• Audio and video speech-to-text from common media formats.
• Public media URL transcription for direct media links and supported public media pages.
• Public media-page workflows where anonymous access is supported by the resolver; source-site availability can change.
• Timestamped transcript segments for downstream search, review, editing, and automation.
• SRT and VTT subtitles generated from the timestamped transcript.
• QA-gated delivery designed to hold suspicious Whisper loops or structurally unreliable outputs.
• Hardware-accelerated processing through ComIQ's NPU / GPU / CPU execution stack.
• Local fulfillment engine — no external speech-to-text API is used by ComIQ's transcription processor.
• Structured Apify Dataset output ready for API use and automation.
• Pay per started audio minute after successful QA-approved delivery.

Quick start

Option A — upload a file

1. Open the Actor.
2. Upload one or more audio/video files.
3. If an upload is stored in one of your existing Apify key-value stores, select the same store under Source storage(s) for uploaded media.
4. Set a suitable Maximum cost per run.
5. Start the run.
6. Open Output / Dataset when the run completes.

Option B — paste a public media URL

1. Paste one or more HTTPS URLs into Public media URLs.
2. No source-storage selection is needed for URLs.
3. Set a suitable Maximum cost per run.
4. Start the run.
5. ComIQ resolves supported public media pages, stages the media temporarily in the Actor run's own storage, transcribes it, and removes the temporary staged file after processing.
6. Use the full transcript, timestamped segments, SRT, VTT, and source metadata from the Dataset.

Supported input

Upload 1–5 media files per run.

Common supported containers/codecs include:

• WAV
• MP3
• M4A
• AAC
• FLAC
• OGG
• OPUS
• MP4
• MOV
• MKV
• WEBM
• AVI
• MPG
• MPEG

ComIQ is designed primarily for spoken audio.

Best use cases

• meeting transcription;
• interview transcription;
• podcast transcription;
• voice-note transcription;
• lecture and presentation transcription;
• recorded customer-call transcription;
• video speech-to-text;
• SRT subtitle generation;
• VTT subtitle generation;
• transcript ingestion for search, RAG, summarization, analytics, and automation.

Not the primary target

Singing and music lyric transcription are not the main use case. Whisper Base can be considerably less reliable on sung vocals than on normal speech.

Search-friendly use cases

Use ComIQ for audio transcription, video transcription, speech to text, audio URL to text, video URL to transcript, public media URL transcription, URL to transcript, MP3 transcription, MP4 transcription, Whisper transcription, meeting transcription, podcast transcription, timestamped transcripts, SRT generation, and VTT subtitle generation.

Public media URL support

Alpha 4.3 adds a second input path for customers who do not want to manually download and re-upload a recording.

Supported URL modes:

• direct HTTPS audio/video file URLs, downloaded inside the Actor and staged temporarily in the run's own storage;
• supported public media-page URLs resolved with yt-dlp;
• public video/audio pages such as Archive.org-style media pages and other sources supported anonymously by the resolver.

ComIQ prefers a single audio stream when resolving a media page because transcription does not require the video track.

URL safety limits

For resolved public media pages, the default limits are:

• 180 minutes maximum duration;
• 200 MB maximum staged download size;
• up to 5 total inputs per run, counting uploads and URLs together.

These limits are configurable in the Actor input within the safe range exposed by the UI.

Source-site compatibility

Public media-page support depends on anonymous access being available from the source site at run time. Some sites can require login, cookies, browser challenges, or other authentication and are therefore not supported by ComIQ's public Store Actor when anonymous access is unavailable. ComIQ does not request customer account credentials or attempt to bypass access controls.

URL limitations

Public media URL support is best-effort because source websites change their delivery systems.

The following are intentionally not supported:

• DRM-protected media;
• private or login-only media;
• credential/cookie bypass;
• live or upcoming streams;
• multi-item playlist processing: if a supported page resolves as a collection/container, ComIQ processes the first media item only;
• media the user is not authorized to process.

A source site may also temporarily block automated access or change in a way that requires a future resolver update.

Output

Each QA-approved transcript is returned as a Dataset item.

Typical fields include:

Field	Description
source_file	Original or resolved media filename
source_kind	Upload, direct URL, or resolved public media page
source_url	Original public media URL when URL input was used
source_title	Resolved public media title when available
source_site	Resolved source service/site when available
language	Detected transcript language
duration_seconds	Media duration
billed_minutes	Started audio minutes billed for the delivered result
text	Full transcript
segments	Timestamped transcript segments
srt	Generated SRT subtitle text
vtt	Generated VTT subtitle text
word_count	Transcript word count
qa_score	ComIQ heuristic QA score
qa_issues	QA warnings detected by ComIQ
actual_device	CPU, GPU, or NPU used for local inference
model	Transcription model identifier
processing_seconds	Local transcription processing time
realtime_factor	Media duration divided by processing time
delivery_status	Result delivery state
billing_mode	Preview or Apify Pay Per Event mode
billing_status	Billing state associated with the delivered result

Example output

{
  "source_file": "meeting.mp4",
  "language": "en",
  "duration_seconds": 618.427,
  "billed_minutes": 11,
  "text": "Full transcript text...",
  "segments": [
    {
      "start": 0.0,
      "end": 2.4,
      "text": "Good morning, and welcome to the meeting."
    }
  ],
  "srt": "1\n00:00:00,000 --> 00:00:02,400\nGood morning, and welcome to the meeting.",
  "vtt": "WEBVTT\n\n00:00:00.000 --> 00:00:02.400\nGood morning, and welcome to the meeting.",
  "qa_score": 100,
  "actual_device": "NPU",
  "processing_seconds": 34.8,
  "realtime_factor": 17.8,
  "delivery_status": "completed",
  "billing_mode": "apify_pay_per_event",
  "billing_status": "charged"
}

QA-held jobs and billing

ComIQ charges transcription-minute events only after a transcript passes the delivery QA gate.

If processing completes but the transcript is held because it is not customer-ready—for example, speech is too sparse, repeated/hallucinated text is detected, or the media is primarily music—the run can complete without a delivered Dataset item, but zero transcription-minute events are charged for that held result.

The small Apify Actor-start/platform cost, if applicable to the customer's Apify plan, is separate from ComIQ's transcription-minute event pricing.

Pricing

ComIQ uses Pay Per Event pricing.

Event	Launch price
Transcription Minute	$0.015 per started audio minute successfully delivered
Actor Start	Small Apify platform event

The primary pricing event is Transcription Minute.

Pricing examples

A successfully delivered transcript is billed by started audio minute:

• 1:00 audio → 1 event → $0.015
• 5:00 audio → 5 events → $0.075
• 10:18 audio → 11 events → $0.165
• 30:00 audio → 30 events → $0.450
• 60:00 audio → 60 events → $0.900

The small Apify Actor Start event is additional.

QA-gated billing

ComIQ's custom transcription-minute billing is tied to customer-ready delivery.

If a result is held by the ComIQ QA gate and no transcript is delivered to the Dataset, ComIQ transcription-minute events are not charged for that held result.

Apify's own platform events are controlled by Apify and may apply independently.

QA-gated delivery

ComIQ screens transcripts before automatic delivery.

The QA system looks for structural signals associated with unreliable speech-to-text output, including:

• repeated or looping adjacent segments;
• suspicious clustered phrase repetition;
• unusually sparse recognition;
• extreme token concentration;
• timestamp progression anomalies;
• other patterns associated with Whisper hallucination or loop behavior.

Important: QA score is not word accuracy

A ComIQ QA score of 100/100 does not mean the transcript is 100% word-for-word accurate.

The QA score measures structural reliability and hallucination/loop risk. Proper nouns, brand names, uncommon terminology, numbers, accents, noisy recordings, overlapping speakers, and poor microphone quality can still cause recognition errors even when QA passes.

For legal, medical, financial, regulatory, safety-critical, or irreversible decisions, review the transcript against the original recording before relying on it.

Performance

ComIQ runs transcription on the connected fulfillment hardware and can use:

1. NPU
2. GPU
3. CPU fallback

On the current validated ComIQ hardware, a 618-second benchmark has completed on the NPU in roughly 29–35 seconds in recent tests, or around 18×–21× realtime.

These figures are examples, not performance guarantees. Actual speed depends on file format, media length, hardware availability, model state, concurrent system load, and source characteristics.

Limited permissions and scoped file access

Alpha 4.2 is designed to support Limited permissions.

When audio or video is uploaded into Apify key-value storage, select the key-value store containing that upload under Source storage(s) for uploaded media. The input schema requests scoped READ + WRITE permission for the selected store(s). Apify can then scope the run token to those specific resources rather than allowing access to unrelated account data.

If you provide an external HTTPS media URL or an already-signed file URL, no Apify source-storage permission is required.

Why WRITE is requested for the selected source store

Apify's key-value-store API only exposes the URL-signing secret to clients that have WRITE permission on that specific store. ComIQ needs that signing capability to create a pre-signed public URL that the external Windows fulfillment worker can download without receiving the user's Apify API token.

The Actor remains Limited permissions overall, and the extra permission is scoped only to the store(s) the user explicitly selects. The current ComIQ code does not intentionally modify or delete the uploaded source records.

The migration code remains compatible with Full-permission runs during validation, but ComIQ's target Store configuration is Limited permissions once the scoped-storage test passes.

Security and processing architecture

• Actor-to-ComIQ job communication uses authenticated HTTPS.
• ComIQ's Windows fulfillment worker uses outbound HTTPS.
• No public inbound port on the processing PC is required for normal fulfillment.
• Transcription is processed by the local ComIQ OpenVINO Whisper pipeline.
• Public media-page URLs and direct external media URLs are fetched in the Apify Actor, temporarily staged in the run's own KVS, then deleted after the ComIQ job finishes.
• Results are returned to the requesting Apify run.
• Do not upload media you are not authorized to process.

Users should determine whether this architecture is appropriate before submitting confidential, regulated, privileged, or highly sensitive recordings.

API and automation

Because this is an Apify Actor, it can be used through:

• Apify Console;
• Apify API;
• Actor Tasks;
• schedules;
• supported webhooks and integrations;
• downstream workflows that consume Dataset output.

Typical automation patterns include:

• meeting recording → transcript → summary;
• podcast → transcript → content workflow;
• video → SRT/VTT subtitles;
• voice note → structured text;
• recorded call → searchable transcript;
• transcript → RAG / semantic search / analytics.

Limitations

Transcription quality may be reduced by:

• overlapping speakers;
• heavy background noise;
• quiet or distorted speech;
• strong music under speech;
• uncommon names and terminology;
• code-switching between languages;
• very strong accents;
• poor source recordings;
• singing or melodic vocals.

The current local model is Whisper Base OpenVINO. ComIQ may introduce additional quality/model tiers after they are benchmarked for speed, accuracy, and commercial economics.

FAQ

What does ComIQ transcribe?

The current product is designed for spoken audio and video.

Does it support MPG/MPEG video?

Yes. Alpha 4.3.3 adds .mpg, .mpeg, and video/mpeg compatibility across the Apify resolver and the Windows fulfillment worker. PyAV/FFmpeg handles the actual decoding.

Does it support MP3 and MP4?

Yes. MP3 audio and MP4 video are among the common supported formats.

Can I paste a YouTube or other public media URL?

Yes, for supported public media pages. Alpha 4.3 resolves the source with yt-dlp, stages the selected media temporarily in the Actor run's own storage, and sends it to ComIQ for transcription. Source-site availability can change and some URLs may be blocked, private, DRM-protected, or otherwise unsupported.

Does ComIQ download an entire video when only transcription is needed?

For public media pages, ComIQ prefers a single audio stream where the source exposes one. Direct media URLs are also downloaded by the Actor and temporarily staged in its own run storage before the external ComIQ worker accesses them. This avoids many hotlink/User-Agent restrictions that can affect direct cross-origin downloads.

Does it generate subtitles?

Yes. Delivered results include timestamped segments plus generated SRT and VTT text.

How many files can I upload?

The current input supports up to 5 media files per run.

How is billing calculated?

Each successfully delivered file is billed by started audio minute at the current Transcription Minute event price.

Is a partial minute rounded up?

Yes. For example, a 10 minute 18 second recording uses 11 Transcription Minute events.

What happens when QA holds a transcript?

A held result is not delivered as a customer-ready Dataset transcript and ComIQ transcription-minute events are not charged for that held result.

Is QA 100/100 the same as 100% transcription accuracy?

No. QA measures structural quality and hallucination/loop risk. It is not a word-error-rate or ground-truth accuracy percentage.

Does ComIQ use an external speech API?

The current ComIQ fulfillment engine uses its local OpenVINO Whisper processor rather than an external speech-to-text API.

Why is there a Source storage(s) field?

Under Limited permissions, the Actor cannot read arbitrary user storage. If you upload media into Apify storage, selecting that key-value store explicitly grants the Actor scoped READ + WRITE access only to that selected resource for the run. External/signed HTTPS URLs do not need this selection.

Does ComIQ need Full permissions?

Alpha 4.2 is designed to migrate the Actor to Limited permissions after the scoped-storage validation succeeds.

What happens if a page resolves as a playlist or collection?

Some legitimate media pages, especially Archive.org-style item pages, are exposed by media resolvers as a collection because the page contains several renditions or media files. ComIQ processes only the first resolved media item for each supplied URL. It does not automatically process an entire playlist, which prevents unexpected multi-item charges.

Why does ComIQ stage direct URLs instead of giving them directly to the Windows worker?

Some public hosts reject cross-origin hotlink requests or generic HTTP clients even when the media URL is publicly reachable in a browser. Alpha 4.3.1 downloads the direct media inside the Apify Actor with an explicit ComIQ User-Agent, stages it temporarily in the Actor run's own storage, and gives the Windows fulfillment worker a signed Apify URL. The staged record is deleted after the job finishes.

Can I retrieve the transcript through an API?

Yes. Delivered results are available in the Actor's Apify Dataset and can be consumed through Apify's API and integrations.

Troubleshooting

If a run does not produce the expected transcript:

1. Check the Actor Log.
2. Confirm the media format is supported.
3. Confirm the recording contains audible speech.
4. Try a clearer source file if the audio is noisy or heavily compressed.
5. Review qa_issues if the result was held.
6. Include the Apify Run ID when reporting a reproducible issue.

About ComIQ

ComIQ is an autonomous paid-work fulfillment engine designed to route marketplace jobs into efficient local CPU, GPU, and NPU processing while applying QA, revenue protection, settlement tracking, power protection, and foreground-use safeguards.

This Store product: ComIQ Audio & Video Transcriber.