Get a quote

POST/v1/jobs/quote

Parameter	Type	Required	Description
type	string	Required	`separation`, `multi_mic_separation`, `target_speaker_extraction`, `music_separation` or `noise_removal`.
inputs	string[]	Required	Upload ids, in order.
range	object	Optional	`{start_seconds, end_seconds}`: price one stretch, as the submission would.

Field	Type	Description
balance_usd	float \| null	`null` for an organization key without the billing permission. `sufficient_credit` still answers.

bash

curl -X POST https://dueta.ai/v1/jobs/quote \
  -H "Authorization: Bearer $DUETA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"type": "separation", "inputs": ["upl_9f2c7a41d0e8"]}'

Response

json

{
  "type": "separation",
  "inputs": ["upl_9f2c7a41d0e8"],
  "billable_seconds": 612.4,
  "total_duration_seconds": 612.4,
  "estimated_cost_usd": 5.1033,
  "balance_usd": 20.0,
  "sufficient_credit": true,
  "can_submit": true,
  "reasons": []
}

// Not submittable yet: every refusal comes back as a sentence, not an error.
{ "sufficient_credit": false, "can_submit": false,
  "reasons": ["These uploads have not finished: upl_4a81be03c5f7",
              "This job costs more than your balance. Add credit in the console."] }

Run part of a recording

Rule	Value
`start_seconds`, `end_seconds`	Number · required · seconds from the start
Offered on	A recording over 30s
Length	30s or more · `end_seconds` minus `start_seconds`
Bounds	0 ≤ `start_seconds` < `end_seconds` ≤ recording length
Whole recording	Same as no `range`
Billed	The range length
Multi Mic Separation	One range cuts every microphone
Target Speaker Extraction	Cuts the recording, never the voice sample
Music Source Separation, `source_url`	With its `probe_id`, where `GET /v1/limits` has `source_url.range` true · checked against the probed length
Multipart form	`range_start_seconds` and `range_end_seconds` · both or neither
422	input_range_invalid
503	input_range_busy · retry after the `Retry-After` seconds
409	submission_collided

bash

# Run seconds 62.5 to 140 only, then price the same stretch.
curl -X POST https://dueta.ai/v1/jobs/noise-remover \
  -H "Authorization: Bearer $DUETA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"inputs": ["upl_9f2c7a41d0e8"],
       "range": {"start_seconds": 62.5, "end_seconds": 140}}'

curl -X POST https://dueta.ai/v1/jobs/quote \
  -H "Authorization: Bearer $DUETA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"type": "noise_removal", "inputs": ["upl_9f2c7a41d0e8"],
       "range": {"start_seconds": 62.5, "end_seconds": 140}}'

Run Music Source Separation

Split music into an audio file for each track.

POST/v1/jobs/music-separation

Parameter	Type	Required	Description
inputs	string[1]	Optional	The song. One of `inputs`, `source_url`.
source_url	string	Optional	A YouTube link; the audio is fetched into an upload of its own. `youtube.com/watch?v=`, `youtu.be/`, `music.youtube.com`, `m.youtube.com`, `shorts/`.
range	object	Optional	`{start_seconds, end_seconds}`. On a song over 30 s, an upload, or a `source_url` with its `probe_id` where `source_url.range` is true: the stretch to run and bill, 30 s or more. Default: the whole song. `input_range_invalid` otherwise.
callback_url	string	Optional	Callbacks.

Limit	Value
Song	Up to 15 min · a file, or a YouTube link
Quote, link	15 min held · `warnings[].code` `source_length_unknown` · charged for the real length
Hold, link	At least 1 min ($1) without `probe_id` · probe first for the exact price
Input	Ideal 44.1 kHz or higher · up to 192 kHz · stereo kept
Output	14 stems · WAV · 44.1 kHz · stereo · same length
Stems	kick · snare_clap · hihat · ride · crash · perc · sub_bass · mid_bass · mid_lead · pad · pluck · lead_vocal · backing_vocal · fx_other
Price	$1/min of song · a 3-min song is $3

bash

curl -X POST https://dueta.ai/v1/jobs/music-separation \
  -H "Authorization: Bearer $DUETA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"inputs": ["upl_9f2c7a41d0e8"]}'
# -> 201, the job. result.stems lists the 14 stems once it succeeds.
# From a YouTube link instead of an upload:
#   -d '{"source_url": "https://www.youtube.com/watch?v=dQw4w9WgXcQ"}'

Run Duplex Conversation Separation

Split a two-person conversation into a WAV file for each speaker.

POST/v1/jobs/separation

Parameter	Type	Required	Description
inputs	string[1]	Required	Exactly one upload id, up to 30 min.
pipeline	object	Optional	A missing key keeps its default.
range	object	Optional	`{start_seconds, end_seconds}`. On a recording over 30 s: the stretch to run and bill, 30 s or more; default: the whole recording. `input_range_invalid` otherwise.
callback_url	string	Optional	Callbacks.

pipeline key	Default	Effect
vocal_separation	true	Separates only the voices from the uploaded audio and removes background music and noise. Turn it off when the recording is already dry speech: a studio double-ender, a headset call, a meeting-room capture with nothing playing.
enhancement	true	Removes the noise left on each track so the speech is clearer. Turn it off on recordings that are already clean, where there is nothing to remove and denoising can only soften the voice.
super_resolution	true	Turns the 16 kHz result into 48 kHz audio for better sound quality. Turn it off when your pipeline reads 16 kHz audio and would throw the extra bandwidth away, as speech recognition and telephony do.
scoring	true	Our own model estimates each track's SI-SDR from the separated audio alone and adds it to the result. Turn it off when you never read the number, such as an automated pipeline that keeps whatever comes back.

Audio	Value
Input	Ideal 44.1 kHz or higher · up to 192 kHz · mono; stereo is averaged
Output	WAV · 48 kHz with `super_resolution` · 16 kHz without
Price	$0.50/min of recording

bash

curl -X POST https://dueta.ai/v1/jobs/separation \
  -H "Authorization: Bearer $DUETA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"inputs": ["upl_9f2c7a41d0e8"],
       "pipeline": {"super_resolution": false}}'
# -> 201, the job. Omit "pipeline" to run every step.

Run Multi Mic Separation

Remove other voices so each mic keeps only its speaker.

POST/v1/jobs/multi-mic-separation

Jobs since 2026-10-16 report type multi_mic_separation; /v1/jobs/dominant-separation and dominant_separation stay accepted forever.

Parameter	Type	Required	Description
inputs	string[2..10]	Required	One per microphone, in alignment order.
stages	object	Optional	A missing key keeps its default.
range	object	Optional	`{start_seconds, end_seconds}`. On microphones over 30 s: one stretch cut from every microphone, inside the shortest, 30 s or more; default: the whole take. `input_range_invalid` otherwise.
callback_url	string	Optional	Callbacks.

stages key	Default	Effect
sync	true	Aligns the recordings from every microphone to the same moment. Turn it off when every recording came off one device that started them together, and is already sample-aligned.
enhance	true	Removes the noise left on each track so the speech is clearer. Turn it off on recordings that are already clean, where there is nothing to remove and denoising can only soften the voice.
bwe	true	Turns the 16 kHz result into 48 kHz audio for better sound quality. Turn it off when your pipeline reads 16 kHz audio and would throw the extra bandwidth away, as speech recognition and telephony do.

Audio	Value
Input	Ideal 16 kHz or higher · up to 192 kHz · one mono recording per microphone
Output	WAV · 48 kHz with `bwe` · 16 kHz without · `result.sample_rate`
Price	$0.50/min of each microphone, summed

bash

# 2-10 upload ids, one per microphone. The FIRST is the timeline the
# others are aligned against. A step you do not name keeps its default.
curl -X POST https://dueta.ai/v1/jobs/multi-mic-separation \
  -H "Authorization: Bearer $DUETA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"inputs": ["upl_9f2c7a41d0e8", "upl_4a81be03c5f7"],
       "stages": {"bwe": false}}'

Run Target Speaker Extraction

Extract only the person you want from a short voice sample.

POST/v1/jobs/target-speaker-extraction

Parameter	Type	Required	Description
inputs	string[2]	Required	The recording, then the target voice.
range	object	Optional	`{start_seconds, end_seconds}`. On a recording over 30 s: the stretch of the recording to run and bill, 30 s or more; the voice sample is never cut, and the default is the whole recording. `input_range_invalid` otherwise.
callback_url	string	Optional	Callbacks.

Limit	Value
Recording	Up to 120 min
Target voice	3 s of speech, 300 s max · 10 s or more recommended
Input	Ideal 16 kHz or higher · up to 192 kHz · mono; stereo is averaged
Output	`target` · WAV · 16 kHz · same length · no bandwidth extension
Price	$0.50/min of recording · target voice free

bash

AUTH="Authorization: Bearer $DUETA_API_KEY"

upload() {  # declare, send the bytes, print the upload id; fails unless they landed
  ID=$(curl -sS -X POST -H "$AUTH" -H "Content-Type: application/json" \
    -d "{\"filename\": \"$1\", \"size_bytes\": $(wc -c < "$1")}" \
    https://dueta.ai/v1/uploads | jq -er .id) || { echo "upload refused: $1" >&2; return 1; }
  curl -fsS -X PUT -H "$AUTH" -H "Content-Type: application/octet-stream" \
    --data-binary @"$1" https://dueta.ai/v1/uploads/$ID/content |
    jq -e '.state == "ready"' > /dev/null || { echo "upload failed: $1" >&2; return 1; }
  echo "$ID"
}
MIX=$(upload conversation.wav) || exit 1   # inputs[0]: the recording
VOICE=$(upload voice.wav) || exit 1        # inputs[1]: the target voice, 3-300 s

JOB=$(curl -sS -X POST -H "$AUTH" -H "Content-Type: application/json" \
  -d "{\"inputs\": [\"$MIX\", \"$VOICE\"]}" \
  https://dueta.ai/v1/jobs/target-speaker-extraction)
JOB_ID=$(echo "$JOB" | jq -er .id) || { echo "$JOB"; exit 1; }   # an error body (credit, limits) has no id

while :; do
  JOB=$(curl -sS -H "$AUTH" https://dueta.ai/v1/jobs/$JOB_ID)
  case "$(echo "$JOB" | jq -r .status)" in
    succeeded) break ;;
    queued|running) sleep 5 ;;
    failed|canceled) echo "$JOB" | jq -r .error_code; exit 1 ;;
    *) echo "$JOB"; exit 1 ;;   # not a job at all: an error body
  esac
done
curl -fsSL -H "$AUTH" -o target.wav https://dueta.ai/v1/jobs/$JOB_ID/stems/target

Run Noise Reduction

Remove background noise and keep the voice clear.

POST/v1/jobs/noise-remover

Parameter	Type	Required	Description
inputs	string[1]	Required	The recording.
range	object	Optional	`{start_seconds, end_seconds}`. On a recording over 30 s: the stretch to run and bill, 30 s or more; default: the whole recording. `input_range_invalid` otherwise.
callback_url	string	Optional	Callbacks.

Limit	Value
Recording	Up to 120 min
Input	up to 192 kHz · mono or stereo; stereo is averaged
Output	`clean` · WAV · mono · same length · the input's rate, 16 kHz at least
Rate	Over 16 kHz: the band above 8 kHz is extended by a model · `result.processing.bandwidth_extended` · `result.sample_rate` is the delivered rate
Level	The input's, within 1 dB · peak under −0.1 dBFS
Output level	Below the audibility threshold: `failed`, `silent_output`, not charged
Price	$0.50/min of recording

bash

curl -X POST https://dueta.ai/v1/jobs/noise-remover \
  -H "Authorization: Bearer $DUETA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"inputs": ["upl_9f2c7a41d0e8"]}'
# -> 201, the job. result.stems is [{"name": "clean"}] once it succeeds;
#    GET /v1/jobs/$JOB_ID/stems/clean is the WAV.

Read the job back

GET/v1/jobs/{id}

Field	Type	Description
id	string	The job id.
status	string	`queued` at creation. Job status.
inputs[]	object[]	One record per recording. `role`: `mixture` or `enrollment` on Target Speaker Extraction, else null.
pipeline	object	Duplex Conversation Separation only. Empty on a Multi Mic Separation job.
result.processing	object	What ran. Prefer it over `stages`.
stages	object	Multi Mic Separation only. What was asked for.
estimated_cost_seconds	float	Input seconds billed: every input summed, Target Speaker Extraction the recording only. Each model's own rate per input minute: `rate_usd_per_minute` in `GET /v1/limits`; rates may differ by model.
estimated_cost_usd	float	Held now, charged on success.
paid_by	object	`{"kind": "personal"}`, or `organization` with `slug`, `name`.
callback_secret	string	Creation response only.

bash

curl -H "Authorization: Bearer $DUETA_API_KEY" https://dueta.ai/v1/jobs/$JOB_ID |
  jq '{status, stage, progress, queue_position, estimated_cost_usd, pipeline}'

Response

json

{
  "id": "0f9c1a7e-4b21-4d5a-9d0e-2c8b6a1f3e77",
  "type": "separation",
  "status": "queued",
  "stage": "queued",
  "progress": 0,
  "eta_seconds": null,
  "eta_range_seconds": null,
  "eta_updated_at": null,
  "delay_reason": null,
  "queue_position": 2,
  "queue_wait_seconds": 108.2,
  "worker_alive": true,
  "heartbeat_age_seconds": 3.2,
  "gpu_degraded": false,
  "pipeline": { "vocal_separation": true, "enhancement": true,
                "super_resolution": false, "scoring": true },
  "inputs": [
    { "index": 0, "upload_id": "upl_9f2c7a41d0e8", "filename": "conversation.wav",
      "duration_seconds": 612.4, "stem": "conversation-00", "sample_rate": 48000,
      "channels": 2, "state": "ready", "checksum_sha256": "b1946ac9..." }
  ],
  "steps": [
    { "name": "vocal_separation", "state": "pending", "progress": 0 },
    { "name": "separation",       "state": "pending", "progress": 0 },
    { "name": "enhancement",      "state": "pending", "progress": 0 },
    { "name": "super_resolution", "state": "skipped", "progress": 0 },
    { "name": "si_sdr",           "state": "pending", "progress": 0 }
  ],
  "estimated_cost_seconds": 612.4,
  "estimated_cost_usd": 5.1033,
  "billed_seconds": null,
  "billed_usd": null,
  "callback_secret": null,
  "error": null,
  "result": null,
  "created_at": "2026-08-28T09:21:44Z",
  "started_at": null,
  "finished_at": null
}

Cases

402: Insufficient credit. Nothing charged.
404: Unknown input id.
409: Input not ready, or email unverified.
400 count: 1 file for separation, music-separation and noise-remover, 2–10 for multi-mic-separation, 2 for target-speaker-extraction.
400 duration: Over 120 min, summed.
400 length: Duplex Conversation Separation over 30 min, Music Source Separation over 15 min.
400 take: Longest over 2× the shortest.
400 body: pipeline values not booleans.
422 step: Unknown stages key, or separate off.
422 field: stages on Duplex Conversation Separation.
422 enrollment: enrollment_is_mixture, enrollment_too_short, enrollment_too_long. Nothing held.
415: Video part.
429: 1 job (4 paid), 20 a minute, 20 unbilled/hour.
503: Queue unreachable. Back off.
Idempotency-Key: Replays the first answer, with Idempotent-Replayed: true. Also on POST /v1/uploads. 24 h; 255 characters.
422 key: Different body; JSON matches by content, not key order.
409 key: Still running. Wait Retry-After.
Retry: Same inputs, order and steps while unfinished: that job, 201.
Retry scope: Not multipart or url.
Retry callback: Kept if one exists.
Re-run: Finished: a new, charged job.

Get a quote

POST/v1/jobs/quote

Parameter	Type	Required	Description
type	string	Required	`separation`, `multi_mic_separation`, `target_speaker_extraction`, `music_separation` or `noise_removal`.
inputs	string[]	Required	Upload ids, in order.
range	object	Optional	`{start_seconds, end_seconds}`: price one stretch, as the submission would.

Field	Type	Description
balance_usd	float \| null	`null` for an organization key without the billing permission. `sufficient_credit` still answers.

bash

curl -X POST https://dueta.ai/v1/jobs/quote \
  -H "Authorization: Bearer $DUETA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"type": "separation", "inputs": ["upl_9f2c7a41d0e8"]}'

Response

json

{
  "type": "separation",
  "inputs": ["upl_9f2c7a41d0e8"],
  "billable_seconds": 612.4,
  "total_duration_seconds": 612.4,
  "estimated_cost_usd": 5.1033,
  "balance_usd": 20.0,
  "sufficient_credit": true,
  "can_submit": true,
  "reasons": []
}

// Not submittable yet: every refusal comes back as a sentence, not an error.
{ "sufficient_credit": false, "can_submit": false,
  "reasons": ["These uploads have not finished: upl_4a81be03c5f7",
              "This job costs more than your balance. Add credit in the console."] }

Run part of a recording

Rule	Value
`start_seconds`, `end_seconds`	Number · required · seconds from the start
Offered on	A recording over 30s
Length	30s or more · `end_seconds` minus `start_seconds`
Bounds	0 ≤ `start_seconds` < `end_seconds` ≤ recording length
Whole recording	Same as no `range`
Billed	The range length
Multi Mic Separation	One range cuts every microphone
Target Speaker Extraction	Cuts the recording, never the voice sample
Music Source Separation, `source_url`	With its `probe_id`, where `GET /v1/limits` has `source_url.range` true · checked against the probed length
Multipart form	`range_start_seconds` and `range_end_seconds` · both or neither
422	input_range_invalid
503	input_range_busy · retry after the `Retry-After` seconds
409	submission_collided

bash

# Run seconds 62.5 to 140 only, then price the same stretch.
curl -X POST https://dueta.ai/v1/jobs/noise-remover \
  -H "Authorization: Bearer $DUETA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"inputs": ["upl_9f2c7a41d0e8"],
       "range": {"start_seconds": 62.5, "end_seconds": 140}}'

curl -X POST https://dueta.ai/v1/jobs/quote \
  -H "Authorization: Bearer $DUETA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"type": "noise_removal", "inputs": ["upl_9f2c7a41d0e8"],
       "range": {"start_seconds": 62.5, "end_seconds": 140}}'

Run Music Source Separation

Split music into an audio file for each track.

POST/v1/jobs/music-separation

Parameter	Type	Required	Description
inputs	string[1]	Optional	The song. One of `inputs`, `source_url`.
source_url	string	Optional	A YouTube link; the audio is fetched into an upload of its own. `youtube.com/watch?v=`, `youtu.be/`, `music.youtube.com`, `m.youtube.com`, `shorts/`.
range	object	Optional	`{start_seconds, end_seconds}`. On a song over 30 s, an upload, or a `source_url` with its `probe_id` where `source_url.range` is true: the stretch to run and bill, 30 s or more. Default: the whole song. `input_range_invalid` otherwise.
callback_url	string	Optional	Callbacks.

Limit	Value
Song	Up to 15 min · a file, or a YouTube link
Quote, link	15 min held · `warnings[].code` `source_length_unknown` · charged for the real length
Hold, link	At least 1 min ($1) without `probe_id` · probe first for the exact price
Input	Ideal 44.1 kHz or higher · up to 192 kHz · stereo kept
Output	14 stems · WAV · 44.1 kHz · stereo · same length
Stems	kick · snare_clap · hihat · ride · crash · perc · sub_bass · mid_bass · mid_lead · pad · pluck · lead_vocal · backing_vocal · fx_other
Price	$1/min of song · a 3-min song is $3

bash

curl -X POST https://dueta.ai/v1/jobs/music-separation \
  -H "Authorization: Bearer $DUETA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"inputs": ["upl_9f2c7a41d0e8"]}'
# -> 201, the job. result.stems lists the 14 stems once it succeeds.
# From a YouTube link instead of an upload:
#   -d '{"source_url": "https://www.youtube.com/watch?v=dQw4w9WgXcQ"}'

Run Duplex Conversation Separation

Split a two-person conversation into a WAV file for each speaker.

POST/v1/jobs/separation

Parameter	Type	Required	Description
inputs	string[1]	Required	Exactly one upload id, up to 30 min.
pipeline	object	Optional	A missing key keeps its default.
range	object	Optional	`{start_seconds, end_seconds}`. On a recording over 30 s: the stretch to run and bill, 30 s or more; default: the whole recording. `input_range_invalid` otherwise.
callback_url	string	Optional	Callbacks.

pipeline key	Default	Effect
vocal_separation	true	Separates only the voices from the uploaded audio and removes background music and noise. Turn it off when the recording is already dry speech: a studio double-ender, a headset call, a meeting-room capture with nothing playing.
enhancement	true	Removes the noise left on each track so the speech is clearer. Turn it off on recordings that are already clean, where there is nothing to remove and denoising can only soften the voice.
super_resolution	true	Turns the 16 kHz result into 48 kHz audio for better sound quality. Turn it off when your pipeline reads 16 kHz audio and would throw the extra bandwidth away, as speech recognition and telephony do.
scoring	true	Our own model estimates each track's SI-SDR from the separated audio alone and adds it to the result. Turn it off when you never read the number, such as an automated pipeline that keeps whatever comes back.

Audio	Value
Input	Ideal 44.1 kHz or higher · up to 192 kHz · mono; stereo is averaged
Output	WAV · 48 kHz with `super_resolution` · 16 kHz without
Price	$0.50/min of recording

bash

curl -X POST https://dueta.ai/v1/jobs/separation \
  -H "Authorization: Bearer $DUETA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"inputs": ["upl_9f2c7a41d0e8"],
       "pipeline": {"super_resolution": false}}'
# -> 201, the job. Omit "pipeline" to run every step.

Run Multi Mic Separation

Remove other voices so each mic keeps only its speaker.

POST/v1/jobs/multi-mic-separation

Jobs since 2026-10-16 report type multi_mic_separation; /v1/jobs/dominant-separation and dominant_separation stay accepted forever.

Parameter	Type	Required	Description
inputs	string[2..10]	Required	One per microphone, in alignment order.
stages	object	Optional	A missing key keeps its default.
range	object	Optional	`{start_seconds, end_seconds}`. On microphones over 30 s: one stretch cut from every microphone, inside the shortest, 30 s or more; default: the whole take. `input_range_invalid` otherwise.
callback_url	string	Optional	Callbacks.

stages key	Default	Effect
sync	true	Aligns the recordings from every microphone to the same moment. Turn it off when every recording came off one device that started them together, and is already sample-aligned.
enhance	true	Removes the noise left on each track so the speech is clearer. Turn it off on recordings that are already clean, where there is nothing to remove and denoising can only soften the voice.
bwe	true	Turns the 16 kHz result into 48 kHz audio for better sound quality. Turn it off when your pipeline reads 16 kHz audio and would throw the extra bandwidth away, as speech recognition and telephony do.

Audio	Value
Input	Ideal 16 kHz or higher · up to 192 kHz · one mono recording per microphone
Output	WAV · 48 kHz with `bwe` · 16 kHz without · `result.sample_rate`
Price	$0.50/min of each microphone, summed

bash

# 2-10 upload ids, one per microphone. The FIRST is the timeline the
# others are aligned against. A step you do not name keeps its default.
curl -X POST https://dueta.ai/v1/jobs/multi-mic-separation \
  -H "Authorization: Bearer $DUETA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"inputs": ["upl_9f2c7a41d0e8", "upl_4a81be03c5f7"],
       "stages": {"bwe": false}}'

Run Target Speaker Extraction

Extract only the person you want from a short voice sample.

POST/v1/jobs/target-speaker-extraction

Parameter	Type	Required	Description
inputs	string[2]	Required	The recording, then the target voice.
range	object	Optional	`{start_seconds, end_seconds}`. On a recording over 30 s: the stretch of the recording to run and bill, 30 s or more; the voice sample is never cut, and the default is the whole recording. `input_range_invalid` otherwise.
callback_url	string	Optional	Callbacks.

Limit	Value
Recording	Up to 120 min
Target voice	3 s of speech, 300 s max · 10 s or more recommended
Input	Ideal 16 kHz or higher · up to 192 kHz · mono; stereo is averaged
Output	`target` · WAV · 16 kHz · same length · no bandwidth extension
Price	$0.50/min of recording · target voice free

bash

AUTH="Authorization: Bearer $DUETA_API_KEY"

upload() {  # declare, send the bytes, print the upload id; fails unless they landed
  ID=$(curl -sS -X POST -H "$AUTH" -H "Content-Type: application/json" \
    -d "{\"filename\": \"$1\", \"size_bytes\": $(wc -c < "$1")}" \
    https://dueta.ai/v1/uploads | jq -er .id) || { echo "upload refused: $1" >&2; return 1; }
  curl -fsS -X PUT -H "$AUTH" -H "Content-Type: application/octet-stream" \
    --data-binary @"$1" https://dueta.ai/v1/uploads/$ID/content |
    jq -e '.state == "ready"' > /dev/null || { echo "upload failed: $1" >&2; return 1; }
  echo "$ID"
}
MIX=$(upload conversation.wav) || exit 1   # inputs[0]: the recording
VOICE=$(upload voice.wav) || exit 1        # inputs[1]: the target voice, 3-300 s

JOB=$(curl -sS -X POST -H "$AUTH" -H "Content-Type: application/json" \
  -d "{\"inputs\": [\"$MIX\", \"$VOICE\"]}" \
  https://dueta.ai/v1/jobs/target-speaker-extraction)
JOB_ID=$(echo "$JOB" | jq -er .id) || { echo "$JOB"; exit 1; }   # an error body (credit, limits) has no id

while :; do
  JOB=$(curl -sS -H "$AUTH" https://dueta.ai/v1/jobs/$JOB_ID)
  case "$(echo "$JOB" | jq -r .status)" in
    succeeded) break ;;
    queued|running) sleep 5 ;;
    failed|canceled) echo "$JOB" | jq -r .error_code; exit 1 ;;
    *) echo "$JOB"; exit 1 ;;   # not a job at all: an error body
  esac
done
curl -fsSL -H "$AUTH" -o target.wav https://dueta.ai/v1/jobs/$JOB_ID/stems/target

Run Noise Reduction

Remove background noise and keep the voice clear.

POST/v1/jobs/noise-remover

Parameter	Type	Required	Description
inputs	string[1]	Required	The recording.
range	object	Optional	`{start_seconds, end_seconds}`. On a recording over 30 s: the stretch to run and bill, 30 s or more; default: the whole recording. `input_range_invalid` otherwise.
callback_url	string	Optional	Callbacks.

Limit	Value
Recording	Up to 120 min
Input	up to 192 kHz · mono or stereo; stereo is averaged
Output	`clean` · WAV · mono · same length · the input's rate, 16 kHz at least
Rate	Over 16 kHz: the band above 8 kHz is extended by a model · `result.processing.bandwidth_extended` · `result.sample_rate` is the delivered rate
Level	The input's, within 1 dB · peak under −0.1 dBFS
Output level	Below the audibility threshold: `failed`, `silent_output`, not charged
Price	$0.50/min of recording

bash

curl -X POST https://dueta.ai/v1/jobs/noise-remover \
  -H "Authorization: Bearer $DUETA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"inputs": ["upl_9f2c7a41d0e8"]}'
# -> 201, the job. result.stems is [{"name": "clean"}] once it succeeds;
#    GET /v1/jobs/$JOB_ID/stems/clean is the WAV.

Read the job back

GET/v1/jobs/{id}

Field	Type	Description
id	string	The job id.
status	string	`queued` at creation. Job status.
inputs[]	object[]	One record per recording. `role`: `mixture` or `enrollment` on Target Speaker Extraction, else null.
pipeline	object	Duplex Conversation Separation only. Empty on a Multi Mic Separation job.
result.processing	object	What ran. Prefer it over `stages`.
stages	object	Multi Mic Separation only. What was asked for.
estimated_cost_seconds	float	Input seconds billed: every input summed, Target Speaker Extraction the recording only. Each model's own rate per input minute: `rate_usd_per_minute` in `GET /v1/limits`; rates may differ by model.
estimated_cost_usd	float	Held now, charged on success.
paid_by	object	`{"kind": "personal"}`, or `organization` with `slug`, `name`.
callback_secret	string	Creation response only.

bash

curl -H "Authorization: Bearer $DUETA_API_KEY" https://dueta.ai/v1/jobs/$JOB_ID |
  jq '{status, stage, progress, queue_position, estimated_cost_usd, pipeline}'

Response

json

{
  "id": "0f9c1a7e-4b21-4d5a-9d0e-2c8b6a1f3e77",
  "type": "separation",
  "status": "queued",
  "stage": "queued",
  "progress": 0,
  "eta_seconds": null,
  "eta_range_seconds": null,
  "eta_updated_at": null,
  "delay_reason": null,
  "queue_position": 2,
  "queue_wait_seconds": 108.2,
  "worker_alive": true,
  "heartbeat_age_seconds": 3.2,
  "gpu_degraded": false,
  "pipeline": { "vocal_separation": true, "enhancement": true,
                "super_resolution": false, "scoring": true },
  "inputs": [
    { "index": 0, "upload_id": "upl_9f2c7a41d0e8", "filename": "conversation.wav",
      "duration_seconds": 612.4, "stem": "conversation-00", "sample_rate": 48000,
      "channels": 2, "state": "ready", "checksum_sha256": "b1946ac9..." }
  ],
  "steps": [
    { "name": "vocal_separation", "state": "pending", "progress": 0 },
    { "name": "separation",       "state": "pending", "progress": 0 },
    { "name": "enhancement",      "state": "pending", "progress": 0 },
    { "name": "super_resolution", "state": "skipped", "progress": 0 },
    { "name": "si_sdr",           "state": "pending", "progress": 0 }
  ],
  "estimated_cost_seconds": 612.4,
  "estimated_cost_usd": 5.1033,
  "billed_seconds": null,
  "billed_usd": null,
  "callback_secret": null,
  "error": null,
  "result": null,
  "created_at": "2026-08-28T09:21:44Z",
  "started_at": null,
  "finished_at": null
}

Cases

402: Insufficient credit. Nothing charged.
404: Unknown input id.
409: Input not ready, or email unverified.
400 count: 1 file for separation, music-separation and noise-remover, 2–10 for multi-mic-separation, 2 for target-speaker-extraction.
400 duration: Over 120 min, summed.
400 length: Duplex Conversation Separation over 30 min, Music Source Separation over 15 min.
400 take: Longest over 2× the shortest.
400 body: pipeline values not booleans.
422 step: Unknown stages key, or separate off.
422 field: stages on Duplex Conversation Separation.
422 enrollment: enrollment_is_mixture, enrollment_too_short, enrollment_too_long. Nothing held.
415: Video part.
429: 1 job (4 paid), 20 a minute, 20 unbilled/hour.
503: Queue unreachable. Back off.
Idempotency-Key: Replays the first answer, with Idempotent-Replayed: true. Also on POST /v1/uploads. 24 h; 255 characters.
422 key: Different body; JSON matches by content, not key order.
409 key: Still running. Wait Retry-After.
Retry: Same inputs, order and steps while unfinished: that job, 201.
Retry scope: Not multipart or url.
Retry callback: Kept if one exists.
Re-run: Finished: a new, charged job.

Run a separation

Get a quote

Run part of a recording

Run Music Source Separation

Run Duplex Conversation Separation

Run Multi Mic Separation

Run Target Speaker Extraction

Run Noise Reduction

Read the job back

Cases

Run a separation

Get a quote

Run part of a recording

Run Music Source Separation

Run Duplex Conversation Separation

Run Multi Mic Separation

Run Target Speaker Extraction

Run Noise Reduction

Read the job back

Cases