HALOWERK Medien

Takes an audio or video file by URL or inline base64 and returns the transcript: the full text, every word with start and end time and its own confidence, the utterances and paragraphs the recogniser separated, and the detected language with its confidence where no language was specified.

Takes two documents, by URL, inline base64 or raw text, extracts the text of each and compares them. Paragraphs are the unit: the answer lists every block that was removed, added, or kept but reworded, and for reworded blocks it names the individual words that went and came, so a single changed number in a long clause is visible instead of the whole clause being flagged. Alongside the content diff

Uses the existing Deepgram and local media pipeline with one diarized transcription call. Subtitle cues split at speaker changes, pauses, six seconds or 84 characters. Unknown speakers remain null. Requires a lawful recording; submitted audio leaves this server for Deepgram, with the existing provider opt-out disclosure. Audio, transcript and subtitles are not retained. This synchronous request ha

Resamples a grayscale image to remove radial and tangential lens distortion, using the camera matrix and five distortion coefficients you supply — typically from a planar calibration. Dimensions and camera matrix are preserved, so the corrected image drops into an existing pipeline without any change to geometry downstream. Because the output frame is unchanged, strongly distorted corners are push

Solves for the rigid transform between an object's coordinate frame and the camera, given 3D landmark coordinates, where those landmarks appear in the image, and the camera matrix and distortion. Returns the rotation and translation together with the reprojection RMSE, so a caller can tell a well-constrained pose from one that merely converged. This is the standard step for placing a known object

Triangulates 3D positions from corresponding points in two calibrated views using the supplied projection matrices. Degenerate cases are reported honestly: a correspondence whose rays are near-parallel has no finite intersection, and it comes back as null with its position in the finite mask, instead of as an enormous coordinate that silently poisons whatever consumes it. That makes the result saf

Solves the hand-eye problem: given a series of gripper-to-base poses read from the robot and the matching target-to-camera poses seen by the camera, it recovers the rigid transform between camera and gripper that stays constant as the arm moves. This is what lets a robot act on what its camera sees. The poses must be genuinely synchronized — a pairing that is off by one sample produces a confident

Fits a 2D projective transform between two point sets with RANSAC at a caller-chosen reprojection threshold, and returns both the homography and the inlier mask. The mask is what makes this useful beyond a plain fit: bad correspondences are named rather than quietly averaged in, so a caller can re-run detection, drop a frame or raise an alarm. Suitable for document rectification, planar tracking a

Computes dense Farneback optical flow between two consecutive grayscale frames and returns the horizontal and vertical displacement of every pixel. Unlike sparse tracking this covers the whole image, which is what you want for measuring motion in regions with no distinct features to track — smoke, water, crowds, a panning background. Both frames must be the same size and should come from the same

Takes an audio or video file by URL or inline base64 and reports who spoke when. Every turn comes back with its speaker label, start and end time, duration, word count and text, and consecutive words from the same speaker are merged into one turn rather than left as a word list.

Compare Separate provider confidence bands from unknown confidence values. against a bounded caller-supplied previous snapshot and return exact change witnesses. Baseline is caller supplied and is never authenticated as historical server output. Pure bounded timing/source-quote validation runs before payment. Exactly one current motor call, zero baseline provider calls. Full current provider discl

Compare Index subtitle display gaps for caption coverage review. against a bounded caller-supplied previous snapshot and return exact change witnesses. Baseline is caller supplied and is never authenticated as historical server output. Pure bounded timing/source-quote validation runs before payment. Exactly one current motor call, zero baseline provider calls. Full current provider disclosure reta

Compare Timestamped source search for API design discussion passages; literal vocabulary only. against a bounded caller-supplied previous snapshot and return exact change witnesses. Baseline is caller supplied and is never authenticated as historical server output. Pure bounded timing/source-quote validation runs before payment. Exactly one current motor call, zero baseline provider calls. Full cu

Compare Timestamped source search for deployment discussion passages; literal vocabulary only. against a bounded caller-supplied previous snapshot and return exact change witnesses. Baseline is caller supplied and is never authenticated as historical server output. Pure bounded timing/source-quote validation runs before payment. Exactly one current motor call, zero baseline provider calls. Full cu

Compare Timestamped source search for roadmap discussion wording; literal vocabulary only. against a bounded caller-supplied previous snapshot and return exact change witnesses. Baseline is caller supplied and is never authenticated as historical server output. Pure bounded timing/source-quote validation runs before payment. Exactly one current motor call, zero baseline provider calls. Full curren

Timestamped source search for venue planning wording; literal vocabulary only. Reuses exactly one existing qualified transcript-package request. Source audio leaves the server for the existing Deepgram processor; existing provider disclosure and cost/margin retained. Deterministic timestamps/text processing only; no acoustic VAD, speaker identity or specialist accuracy claims. Base motor quote plu

Produce path-free FFmpeg filter expressions for source-backed speaker turns. Reuses exactly one existing qualified transcript-package request. Source audio leaves the server for the existing Deepgram processor; existing provider disclosure and cost/margin retained. Deterministic timestamps/text processing only; no acoustic VAD, speaker identity or specialist accuracy claims. Base motor quote plus0

Propose padded cue clips around caller-specified literal keywords. Reuses exactly one existing qualified transcript-package request. Source audio leaves the server for the existing Deepgram processor; existing provider disclosure and cost/margin retained. Deterministic timestamps/text processing only; no acoustic VAD, speaker identity or specialist accuracy claims. Base motor quote plus0.001USDC;

Compare Propose padded cue clips around caller-specified literal keywords. against a bounded caller-supplied previous snapshot and return exact change witnesses. Baseline is caller supplied and is never authenticated as historical server output. Pure bounded timing/source-quote validation runs before payment. Exactly one current motor call, zero baseline provider calls. Full current provider discl

Compare Locate caller-specified literal redaction strings; no semantic PII detection. against a bounded caller-supplied previous snapshot and return exact change witnesses. Baseline is caller supplied and is never authenticated as historical server output. Pure bounded timing/source-quote validation runs before payment. Exactly one current motor call, zero baseline provider calls. Full current pro

Compare Timestamped source search for loading discussion wording; literal vocabulary only. against a bounded caller-supplied previous snapshot and return exact change witnesses. Baseline is caller supplied and is never authenticated as historical server output. Pure bounded timing/source-quote validation runs before payment. Exactly one current motor call, zero baseline provider calls. Full curren

Timestamped source search for return planning wording; literal vocabulary only. Reuses exactly one existing qualified transcript-package request. Source audio leaves the server for the existing Deepgram processor; existing provider disclosure and cost/margin retained. Deterministic timestamps/text processing only; no acoustic VAD, speaker identity or specialist accuracy claims. Base motor quote pl

Compare Timestamped source search for shortage discussion wording; literal vocabulary only. against a bounded caller-supplied previous snapshot and return exact change witnesses. Baseline is caller supplied and is never authenticated as historical server output. Pure bounded timing/source-quote validation runs before payment. Exactly one current motor call, zero baseline provider calls. Full curre

Compare Timestamped source search for maintenance planning passages; literal vocabulary only. against a bounded caller-supplied previous snapshot and return exact change witnesses. Baseline is caller supplied and is never authenticated as historical server output. Pure bounded timing/source-quote validation runs before payment. Exactly one current motor call, zero baseline provider calls. Full cur

Compare Timestamped source search for consent vocabulary, no legal conclusion; literal vocabulary only. against a bounded caller-supplied previous snapshot and return exact change witnesses. Baseline is caller supplied and is never authenticated as historical server output. Pure bounded timing/source-quote validation runs before payment. Exactly one current motor call, zero baseline provider calls

Timestamped source search for publication planning passages; literal vocabulary only. Reuses exactly one existing qualified transcript-package request. Source audio leaves the server for the existing Deepgram processor; existing provider disclosure and cost/margin retained. Deterministic timestamps/text processing only; no acoustic VAD, speaker identity or specialist accuracy claims. Base motor qu

Compare Timestamped source search for competition wording; literal vocabulary only. against a bounded caller-supplied previous snapshot and return exact change witnesses. Baseline is caller supplied and is never authenticated as historical server output. Pure bounded timing/source-quote validation runs before payment. Exactly one current motor call, zero baseline provider calls. Full current provi

Timestamped source search for procurement passages; literal vocabulary only. Reuses exactly one existing qualified transcript-package request. Source audio leaves the server for the existing Deepgram processor; existing provider disclosure and cost/margin retained. Deterministic timestamps/text processing only; no acoustic VAD, speaker identity or specialist accuracy claims. Base motor quote plus0

Compare Cross-reference subtitle cues with contained source words and flag label mismatch. against a bounded caller-supplied previous snapshot and return exact change witnesses. Baseline is caller supplied and is never authenticated as historical server output. Pure bounded timing/source-quote validation runs before payment. Exactly one current motor call, zero baseline provider calls. Full curren

Compare Flag fast, short and text-heavy cues for explicit editorial review. against a bounded caller-supplied previous snapshot and return exact change witnesses. Baseline is caller supplied and is never authenticated as historical server output. Pure bounded timing/source-quote validation runs before payment. Exactly one current motor call, zero baseline provider calls. Full current provider disc

Timestamped source search for escalation wording; literal vocabulary only. Reuses exactly one existing qualified transcript-package request. Source audio leaves the server for the existing Deepgram processor; existing provider disclosure and cost/margin retained. Deterministic timestamps/text processing only; no acoustic VAD, speaker identity or specialist accuracy claims. Base motor quote plus0.0

Finds the external contours in a caller-supplied binary mask and reports, for each, its area, perimeter, centroid and bounding rectangle, plus the total foreground pixel count. The division of labour is deliberate: segmentation is a domain decision, so the mask comes from you and the measurement comes from here, which keeps the numbers reproducible and independent of any threshold this service mig

Takes matched object and image points from multiple views of a planar calibration target and returns the camera matrix, five distortion coefficients and the RMS reprojection error that the fit actually achieved. The error figure is the point: it tells a caller whether the calibration is usable before anything is built on top of it, instead of handing back numbers that look plausible.

Given paired calibration observations and the intrinsics of both cameras, this computes the rotation and translation between them and derives everything a stereo pipeline needs downstream: rectification transforms for both views, both projection matrices, and the disparity-to-depth mapping. The RMS error in pixels says how well the rig geometry actually fits the data.

Runs semi-global block matching over a rectified grayscale stereo pair and returns per-pixel disparity together with the count of pixels that could actually be matched. Pixels with no reliable match stay unmatched rather than being interpolated, so occlusions and textureless regions are visible in the output instead of hidden by plausible-looking values. Disparities are positive and bounded at 16

Takes an audio or video file and returns a cleaned speech recording. A high-pass filter removes rumble and mains hum, an FFT noise reducer with continuous noise tracking lowers steady background noise such as fans, air conditioning and hiss by 10, 18 or 28 dB, a declicker removes short clicks, and the result is normalised to the target loudness in two passes.

Takes a public URL of an audio or video file of up to 120 minutes and 150 MB and returns the transcript: full text, paragraphs with sentences and timings, utterances, the detected language, and a confidence breakdown naming the weakest words with timestamps; every word with its timing on request. The file is downloaded and its running time measured locally with ffprobe before anything is sent, so

Takes an audio or video file and the text spoken in it, and returns your text with timings: every word with start and end, every line with start and end, and ready SRT, WebVTT and LRC (karaoke/lyrics) files built from your own lines, so your spelling, punctuation and line breaks are kept. The recording is transcribed with word timestamps and your words are laid onto the recognised words by global

Takes an audio or video file and returns a new audio file in which sensitive spoken content is covered by a 1 kHz tone or by silence. The recording is transcribed with word timestamps; then fixed, checkable patterns are searched in the words — e-mail addresses, also in spoken form, phone numbers of 7 to 15 digits, IBANs, card numbers that pass the Luhn check, and optionally any number of four or m

Takes 2 to 15 recordings by URL or inline base64, each with its declared length, and returns one transcript per file in the order given: full text, detected language, paragraphs, utterances, a confidence summary naming the weakest words with timestamps, and optionally every word with its timing. Billing is per started audio minute across the whole batch, at 0.017 USDC instead of the 0.02 USDC of a

Takes an audio or video file by URL or inline base64 and returns subtitles in the target language with the timing of the original speech. The recording is transcribed with word timestamps, split into subtitle cues at pauses, six seconds or 84 characters, and every cue is translated on its own so that the number and timing of cues stay identical — this is checked on every model answer and a mismatc

Takes a video by URL or inline base64 and returns what an agent needs to work with it without watching it. The soundtrack is transcribed with paragraphs and timestamps; scene changes are detected with ffmpeg; chapters are cut at paragraph starts, preferring those within three seconds of a visual cut, at least min_chapter_seconds apart and at most 20.

Takes an audio or video file of up to 60 minutes and returns an edited audio file without filler words and with long pauses shortened. Speech is recognised with filler-word detection; every uh, um, er, äh, ähm, hm and similar is cut at its word boundaries, and every pause longer than max_pause_seconds is shortened to keep_pause_seconds rather than removed, so the speech keeps its rhythm.

Takes 1 to 15 recordings, each with its declared length (together at most 60 minutes), and up to 25 search terms or phrases. All files are transcribed with word timestamps, then every term is searched word by word, ignoring case and punctuation. Each hit carries the file, start and end time and eight words of context on either side.

Takes a recording and the text that was supposed to be read, and returns a reading assessment. The recording is transcribed with word timestamps and aligned to the text by global sequence alignment, ignoring case and punctuation. Every text word is marked correct, substituted (with what was heard), or omitted; words spoken but not in the text are listed as insertions.

Takes an audio or video file of up to 60 minutes and returns the best short clips: segments that stand on their own with a clear opening and a complete thought, between min_seconds and max_seconds long. A language model picks the segments as sentence numbers only; start and end times come from the recogniser and lengths are checked here, so a clip never starts or ends mid-sentence. Each clip carri

Takes a podcast episode of up to 120 minutes by URL or inline and returns publish-ready show notes: a title suggestion, a two-to-four-sentence description, chapters with timestamps (as text block for podcast players and YouTube), the people, organisations, products and links mentioned, up to five quotable sentences and keywords, plus the paragraphed transcript. Chapter boundaries are chosen by a l

Takes a meeting, call or interview recording and returns structured minutes. The audio is transcribed with speaker separation (S1, S2, ...); a language model then extracts the summary, topics, decisions, action items with the responsible person and deadline as they were spoken, and open questions.

Takes a recorded call or conversation of up to 60 minutes and returns conversation metrics computed from word timings and speaker separation: talk time and share per speaker, words per minute, number of turns, interruptions (a speaker starting while the other is still speaking or within 0.2 s), total and longest silence, longest monologue. Required phrases you pass are searched word by word and re

Takes a text such as a chapter, article or report of up to 20,000 characters and returns it read aloud as one audio file. Lines starting with "# " become chapters; blank lines separate paragraphs, and long paragraphs are split at sentence boundaries. Every paragraph is spoken separately by a Deepgram voice of the chosen language (English, German, Spanish, French, Italian, Dutch or Japanese; automa