Voice Cloning & Verification

Clone any voice from a short audio sample and generate speech in 23 languages with full emotion control. Zero-shot voice cloning requires no fine-tuning — upload a 10-30 second reference clip and start generating immediately. Built on FOTOhub's proprietary IDA Voice engine running on dedicated GPU infrastructure.

Capability	Description	Credits
Voice Cloning	Create a reusable voice from a reference audio clip	2
Clone Preview	Test a cloned voice before saving	Free
Emotional TTS	Generate speech with 10-dimensional emotion control	1
Expressive TTS	Tag-based segment generation with inline emotion/SFX	1
Voice Comparison	Compare two audio samples for speaker similarity	1
Deepfake Detection	Detect AI-generated or manipulated audio	2
Voice Forensics	Full forensic analysis report	3

Voice Cloning Workflow

The voice cloning pipeline follows four steps: record, upload, clone, and use.

┌─────────┐     ┌─────────┐     ┌─────────────┐     ┌──────────────┐
│ Record  │ ──► │ Upload  │ ──► │ Clone Voice │ ──► │ Generate TTS │
│ 10-30s  │     │ Audio   │     │ (2 credits) │     │ with voice   │
└─────────┘     └─────────┘     └─────────────┘     └──────────────┘

Record a clear audio sample (10-30 seconds) of the target voice
Upload the audio file via the clone endpoint (WAV, MP3, FLAC, OGG, M4A supported)
Clone the voice — the system extracts speaker embeddings and creates a reusable voice profile
Generate speech using the cloned voice_id in any TTS endpoint

Reference Audio Quality

For best results, use a clean recording with minimal background noise, no music, and a single speaker. Longer samples (20-30s) produce higher fidelity clones. The speaker should use natural speech patterns — avoid reading in a monotone.

Endpoints

Create Voice Clone

Clone a voice from an uploaded audio file. Extracts speaker embeddings via IDA Voice on GPU infrastructure and saves the voice profile for reuse.

POST /v1/ai/voice/clone

Content-Type: multipart/form-data

Parameters

Parameter	Type	Required	Default	Description
`file`	file	Yes	—	Audio file (WAV, MP3, FLAC, OGG, M4A). 10-30s recommended, max 50 MB.
`voice_name`	string	Yes	—	Display name for the cloned voice.
`language`	string	No	`"en"`	Primary language of the reference audio.
`gender`	integer	No	`1`	Speaker gender: `1` = male, `2` = female.
`age`	integer	No	`30`	Approximate speaker age (used for metadata).
`description`	string	No	`""`	Optional description of the voice characteristics.
`enhance_audio`	boolean	No	`false`	Apply noise reduction to the reference before cloning.
`generate_preview`	boolean	No	`true`	Generate a TTS preview sample after cloning.

Response

json

{
  "success": true,
  "voice_id": "vc_a1b2c3d4e5f6",
  "voice_uid": "vc_a1b2c3d4e5f6",
  "name": "Alex Narrator",
  "preview_url": "https://s1.fotohub.app/storage/v1/object/sign/audio/user123/voice-clones/vc_a1b2c3d4e5f6-sample.mp3?token=...",
  "generation_id": "550e8400-e29b-41d4-a716-446655440000",
  "credits_used": 2,
  "processing_time_ms": 4250,
  "sample_duration": 3.2,
  "provider": "ida-voice"
}

Example

PythonTypeScriptcURL

python

from fotohub import FotoHub

client = FotoHub(api_key="fh_live_...")

# Clone a voice from a reference audio file
result = client.voice.clone(
    file=open("reference_voice.wav", "rb"),
    voice_name="Alex Narrator",
    language="en",
    gender=1,
    description="Deep male narrator voice, warm tone"
)

print(f"Voice ID: {result.voice_id}")
print(f"Preview: {result.preview_url}")

typescript

import { FotoHub } from "fotohub";
import fs from "fs";

const client = new FotoHub({ apiKey: "fh_live_..." });

const result = await client.voice.clone({
  file: fs.createReadStream("reference_voice.wav"),
  voiceName: "Alex Narrator",
  language: "en",
  gender: 1,
  description: "Deep male narrator voice, warm tone",
});

console.log(`Voice ID: ${result.voiceId}`);
console.log(`Preview: ${result.previewUrl}`);

bash

curl -X POST "https://apis.fotohub.app/v1/ai/voice/clone" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "file=@reference_voice.wav" \
  -F "voice_name=Alex Narrator" \
  -F "language=en" \
  -F "gender=1" \
  -F "description=Deep male narrator voice, warm tone" \
  -F "generate_preview=true"

Preview Cloned Voice

Generate a TTS preview for a cloned voice without saving. Useful for testing before committing to a clone.

POST /v1/ai/voice/clone/preview

Parameters

Parameter	Type	Required	Default	Description
`voice_uid`	string	Yes	—	The voice ID returned from the clone endpoint.
`language`	string	No	`"en"`	Language for the preview text.
`text`	string	No	auto	Custom preview text. If omitted, uses a built-in sample for the selected language.

Response

json

{
  "success": true,
  "preview_url": "https://s1.fotohub.app/storage/v1/object/sign/audio/user123/voice-previews/vc_a1b2c3d4e5f6_1721654321000.mp3?token=..."
}

Example

PythonTypeScriptcURL

python

result = client.voice.preview(
    voice_uid="vc_a1b2c3d4e5f6",
    language="en",
    text="Hello, this is a test of my cloned voice."
)
print(f"Preview URL: {result.preview_url}")

typescript

const preview = await client.voice.preview({
  voiceUid: "vc_a1b2c3d4e5f6",
  language: "en",
  text: "Hello, this is a test of my cloned voice.",
});
console.log(`Preview URL: ${preview.previewUrl}`);

bash

curl -X POST "https://apis.fotohub.app/v1/ai/voice/clone/preview" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "voice_uid": "vc_a1b2c3d4e5f6",
    "language": "en",
    "text": "Hello, this is a test of my cloned voice."
  }'

List Cloned Voices

Retrieve all saved cloned voices for the authenticated user.

POST /v1/ai/voice/clone/list

Response

json

{
  "success": true,
  "voices": [
    {
      "id": "550e8400-e29b-41d4-a716-446655440000",
      "voice_id": "vc_a1b2c3d4e5f6",
      "voice_uid": "vc_a1b2c3d4e5f6",
      "name": "Alex Narrator",
      "mode": "clone",
      "description": "Deep male narrator voice, warm tone",
      "language": "en",
      "gender": 1,
      "preview_url": "https://s1.fotohub.app/storage/v1/object/sign/audio/...",
      "created_at": "2026-07-20T14:30:00Z",
      "provider": "ida-voice"
    }
  ]
}

Example

PythonTypeScriptcURL

python

voices = client.voice.list()
for voice in voices.voices:
    print(f"{voice.name} ({voice.voice_id}) — {voice.language}")

typescript

const voices = await client.voice.list();
for (const voice of voices.voices) {
  console.log(`${voice.name} (${voice.voiceId}) - ${voice.language}`);
}

bash

curl -X POST "https://apis.fotohub.app/v1/ai/voice/clone/list" \
  -H "Authorization: Bearer YOUR_API_KEY"

Delete Cloned Voice

Permanently delete a cloned voice profile.

DELETE /v1/ai/voice/clone/{voice_uid}

Response

json

{
  "success": true,
  "status": "deleted"
}

Example

PythonTypeScriptcURL

python

client.voice.delete("vc_a1b2c3d4e5f6")

typescript

await client.voice.delete("vc_a1b2c3d4e5f6");

bash

curl -X DELETE "https://apis.fotohub.app/v1/ai/voice/clone/vc_a1b2c3d4e5f6" \
  -H "Authorization: Bearer YOUR_API_KEY"

Emotional TTS

Generate speech with fine-grained emotion control using a 10-dimensional emotion vector. Each dimension represents an emotion intensity from 0.0 to 1.0, and dimensions can be mixed for nuanced expressions.

POST /v1/ai/voice/emotional

Emotion Dimensions

Index	Dimension	Description
0	`happy`	Joy, happiness, delight
1	`sad`	Sadness, sorrow, melancholy
2	`angry`	Anger, fury, frustration
3	`fear`	Fear, anxiety, terror
4	`surprise`	Surprise, astonishment, shock
5	`disgust`	Disgust, revulsion, contempt
6	`calm`	Calm, peaceful, serene
7	`neutral`	Neutral, default, no emotion
8	`whisper`	Whisper, hushed, breathy
9	`shout`	Shout, loud, forceful

Parameters

Parameter	Type	Required	Default	Description
`text`	string	Yes	—	Text to synthesize (max 5000 characters).
`language`	string	No	`"en"`	Output language code (see supported languages).
`emotion_vec`	array	No	neutral	10-element float array, each 0.0-1.0.
`emotion`	string	No	—	Shorthand: emotion name (e.g., `"happy"`, `"angry"`).
`intensity`	float	No	`1.0`	Intensity multiplier when using `emotion` shorthand.
`voice_uid`	string	No	—	Use a cloned voice by its ID.
`builtin_voice`	string	No	—	Use a built-in voice preset.
`cfg_weight`	float	No	`0.5`	Classifier-free guidance weight (0.0-1.0).
`temperature`	float	No	`0.8`	Sampling temperature (0.05-5.0).
`seed`	integer	No	`0`	Reproducibility seed (0 = random).

Response

Returns audio bytes directly with metadata in response headers:

Header	Description
`X-Processing-Time-Ms`	Generation time in milliseconds
`X-Audio-Duration`	Output audio duration in seconds
`X-Sample-Rate`	Audio sample rate (Hz)
`X-Emotion-Vec`	Applied emotion vector (comma-separated)

Example

PythonTypeScriptcURL

python

from fotohub import FotoHub

client = FotoHub(api_key="fh_live_...")

# Generate with mixed emotions: mostly happy with a hint of surprise
audio = client.voice.emotional(
    text="I can't believe we actually won the championship!",
    language="en",
    emotion_vec=[0.7, 0.0, 0.0, 0.0, 0.4, 0.0, 0.0, 0.0, 0.0, 0.0],
    voice_uid="vc_a1b2c3d4e5f6",
)

with open("excited_speech.wav", "wb") as f:
    f.write(audio.content)

typescript

import { FotoHub } from "fotohub";

const client = new FotoHub({ apiKey: "fh_live_..." });

// Generate with emotion shorthand
const audio = await client.voice.emotional({
  text: "I can't believe we actually won the championship!",
  language: "en",
  emotion: "happy",
  intensity: 0.8,
  voiceUid: "vc_a1b2c3d4e5f6",
});

fs.writeFileSync("excited_speech.wav", audio.content);

bash

curl -X POST "https://apis.fotohub.app/v1/ai/voice/emotional" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "I cannot believe we actually won the championship!",
    "language": "en",
    "emotion_vec": [0.7, 0.0, 0.0, 0.0, 0.4, 0.0, 0.0, 0.0, 0.0, 0.0],
    "voice_uid": "vc_a1b2c3d4e5f6"
  }' \
  --output excited_speech.wav

Emotion Mixing

Dimensions are independent and can be combined. For example, [0.5, 0.3, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0] produces a bittersweet tone mixing happiness with sadness. The model normalizes extreme combinations internally.

Expressive TTS

Generate speech with inline emotion tags, pauses, and sound effect insertions. The text is parsed into segments, each generated with mapped parameters and concatenated with crossfade.

POST /v1/ai/generate/speech

Set speech_model: "ida-voice" (IDA Voice) and include tags in the text field.

Supported Tags

Tag Type	Syntax	Example
Emotion	`[emotion_name]`	`[excited]Hello everyone![calm]Let me explain.`
Pause	`<break time="Xms"/>`	`Hello.<break time="500ms"/>How are you?`
Pause (short)	`<#seconds#>`	`Hello.<#0.5#>How are you?`
Sound	`(sound_name)`	`That was hilarious!(laughing)Anyway...`

Available Emotions

High energy: happy, excited, cheerful, enthusiastic, joyful Low energy: sad, melancholy, disappointed Aggressive: angry, furious, shouting, frustrated Soft: whispering, gentle, calm, reassuring, soothing Reactive: fearful, scared, surprised, shocked Stylistic: sarcastic, dramatic, mysterious, romantic, ironic Functional: speaking slowly, speaking fast, patient, teaching, narrating

Available Sound Effects

laughing, chuckling, sighing, gasping, coughing, clearing throat, yawning, crying, applause, sniffing, groaning

Parameters

Parameter	Type	Required	Default	Description
`text`	string	Yes	—	Text with optional inline tags.
`language`	string	No	`"en"`	Output language code.
`voice_uid`	string	No	—	Cloned voice ID.
`builtin_voice`	string	No	—	Built-in voice preset name.
`exaggeration`	float	No	`0.5`	Base expressiveness (0.0-2.0). Higher = more dramatic.
`cfg_weight`	float	No	`0.5`	Classifier-free guidance (0.0-1.0). Higher = more controlled.
`temperature`	float	No	`0.8`	Sampling randomness (0.5-5.0).
`speed`	float	No	`1.0`	Playback speed (0.5-2.0). Post-generation time-stretch.
`crossfade_ms`	integer	No	`50`	Crossfade duration between segments (ms).
`seed`	integer	No	`0`	Reproducibility seed.

Example

PythonTypeScriptcURL

python

from fotohub import FotoHub

client = FotoHub(api_key="fh_live_...")

# Expressive narration with emotion shifts and pauses
script = """[excited]Welcome to the show, everyone!
<break time="800ms"/>
[calm]Today we're going to explore something fascinating.
<#0.5#>
[whispering]But first, let me tell you a secret...
(gasping)
[surprised]Did you hear that?!"""

audio = client.voice.generate(
    text=script,
    language="en",
    voice_uid="vc_a1b2c3d4e5f6",
    exaggeration=0.7,
    speed=1.0,
)

with open("narration.wav", "wb") as f:
    f.write(audio.content)

typescript

import { FotoHub } from "fotohub";

const client = new FotoHub({ apiKey: "fh_live_..." });

const script = `[excited]Welcome to the show, everyone!
<break time="800ms"/>
[calm]Today we're going to explore something fascinating.
<#0.5#>
[whispering]But first, let me tell you a secret...
(gasping)
[surprised]Did you hear that?!`;

const audio = await client.voice.generate({
  text: script,
  language: "en",
  voiceUid: "vc_a1b2c3d4e5f6",
  exaggeration: 0.7,
  speed: 1.0,
});

fs.writeFileSync("narration.wav", audio.content);

bash

curl -X POST "https://apis.fotohub.app/v1/ai/generate/speech" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "[excited]Welcome to the show!(gasping)[surprised]What was that?!",
    "language": "en",
    "speech_model": "ida-voice",
    "voice_uid": "vc_a1b2c3d4e5f6",
    "exaggeration": 0.7
  }' \
  --output narration.wav

Voice Comparison

Compare two audio samples to determine if they contain the same speaker. Uses SpeechBrain ECAPA-TDNN embeddings with state-of-the-art accuracy (0.8% EER on VoxCeleb1).

POST /v1/ai/voice/compare

Content-Type: multipart/form-data

Parameters

Parameter	Type	Required	Default	Description
`reference_audio`	file	Yes	—	Original voice sample (reference).
`test_audio`	file	Yes	—	Voice sample to compare against the reference.
`threshold`	float	No	`0.25`	Similarity threshold. Lower = stricter matching.

Response

json

{
  "success": true,
  "similarity_score": 0.8234,
  "is_same_speaker": true,
  "confidence": "very_high",
  "threshold_used": 0.25,
  "processing_time_ms": 1450,
  "credits_used": 1
}

Confidence levels:

Score Range	Confidence	Interpretation
> 0.70	`very_high`	Almost certainly the same speaker
0.50 - 0.70	`high`	Very likely the same speaker
0.30 - 0.50	`medium`	Possibly the same speaker
< 0.30	`low`	Unlikely to be the same speaker

Example

PythonTypeScriptcURL

python

result = client.voice.compare(
    reference_audio=open("original_voice.wav", "rb"),
    test_audio=open("suspicious_audio.mp3", "rb"),
    threshold=0.25,
)

print(f"Same speaker: {result.is_same_speaker}")
print(f"Similarity: {result.similarity_score:.1%}")
print(f"Confidence: {result.confidence}")

typescript

const result = await client.voice.compare({
  referenceAudio: fs.createReadStream("original_voice.wav"),
  testAudio: fs.createReadStream("suspicious_audio.mp3"),
  threshold: 0.25,
});

console.log(`Same speaker: ${result.isSameSpeaker}`);
console.log(`Similarity: ${(result.similarityScore * 100).toFixed(1)}%`);

bash

curl -X POST "https://apis.fotohub.app/v1/ai/voice/compare" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "reference_audio=@original_voice.wav" \
  -F "test_audio=@suspicious_audio.mp3" \
  -F "threshold=0.25"

Deepfake Detection

Analyze audio to determine if it was generated by AI or synthetically manipulated. Uses a wav2vec2-large model fine-tuned on the ASVspoof 2021 dataset, detecting TTS synthesis, voice conversion, replay attacks, and GAN-generated audio.

POST /v1/ai/voice/deepfake-detect

Content-Type: multipart/form-data

Parameters

Parameter	Type	Required	Default	Description
`audio_file`	file	Yes	—	Audio file to analyze (max 200 MB, max 10 minutes).

Response

json

{
  "success": true,
  "is_deepfake": true,
  "deepfake_probability": 0.9142,
  "confidence": "very_high",
  "model_used": "MelodyMachine/Deepfake-audio-detection-V2",
  "processing_time_ms": 2830,
  "credits_used": 2
}

Probability interpretation:

Range	Assessment
0.85 - 1.00	Almost certainly AI-generated
0.65 - 0.85	Likely AI-generated
0.35 - 0.65	Inconclusive — requires further analysis
0.15 - 0.35	Likely authentic
0.00 - 0.15	Almost certainly authentic

Example

PythonTypeScriptcURL

python

result = client.voice.deepfake_detect(
    audio_file=open("suspicious_call.mp3", "rb")
)

if result.is_deepfake:
    print(f"WARNING: AI-generated audio detected ({result.deepfake_probability:.1%} probability)")
else:
    print(f"Audio appears authentic ({1 - result.deepfake_probability:.1%} confidence)")

typescript

const result = await client.voice.deepfakeDetect({
  audioFile: fs.createReadStream("suspicious_call.mp3"),
});

if (result.isDeepfake) {
  console.log(`WARNING: AI-generated (${(result.deepfakeProbability * 100).toFixed(1)}%)`);
} else {
  console.log(`Audio appears authentic`);
}

bash

curl -X POST "https://apis.fotohub.app/v1/ai/voice/deepfake-detect" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "audio_file=@suspicious_call.mp3"

Voice Forensics

Run a comprehensive forensic analysis on audio including spectral consistency checks, compression artifact detection, splice/edit detection, and noise floor analysis.

POST /v1/ai/voice/forensics

Content-Type: multipart/form-data

Parameters

Parameter	Type	Required	Default	Description
`primary_audio`	file	Yes	—	Main audio file to analyze.
`reference_audio`	file	No	—	Optional reference file for speaker comparison.

Response (Full Report)

json

{
  "success": true,
  "report_id": "550e8400-e29b-41d4-a716-446655440000",
  "speaker_comparison": {
    "similarity_score": 0.7821,
    "is_same_speaker": true,
    "confidence": "very_high"
  },
  "deepfake_analysis": [
    {
      "filename": "interview.mp3",
      "is_deepfake": false,
      "deepfake_probability": 0.0823,
      "confidence": "very_high"
    }
  ],
  "forensic_analysis": [
    {
      "filename": "interview.mp3",
      "duration_seconds": 45.2,
      "authenticity_score": 0.92,
      "spectral_centroid_mean": 1842.3,
      "spectral_consistency": 0.94,
      "compression_detected": false,
      "energy_jumps_count": 1,
      "energy_jump_positions": [12.45],
      "rms_mean_db": -18.4,
      "silence_ratio": 0.08
    }
  ],
  "overall_risk_level": "low",
  "overall_score": 0.92,
  "recommendations": [
    "Audio appears authentic - no anomalies detected"
  ],
  "processing_time_ms": 5420,
  "credits_used": 3
}

Risk levels:

Level	Score	Meaning
`low`	0.8 - 1.0	No significant anomalies detected
`medium`	0.6 - 0.8	Minor inconsistencies, may warrant review
`high`	0.4 - 0.6	Significant anomalies detected
`critical`	0.0 - 0.4	Strong indicators of manipulation or synthesis

Forensic Metrics Explained

Metric	What It Measures
`authenticity_score`	Overall authenticity (0-1, higher = more authentic)
`spectral_consistency`	Whether spectral profile is uniform throughout (detects splices)
`compression_detected`	Whether lossy compression artifacts are present
`energy_jumps_count`	Number of sudden volume changes (indicates edits)
`spectral_flatness_mean`	How noise-like vs tonal the signal is (synthesis indicator)
`silence_ratio`	Proportion of silent frames (unusual patterns = manipulation)

Example

PythonTypeScriptcURL

python

from fotohub import FotoHub

client = FotoHub(api_key="fh_live_...")

# Full forensic analysis with speaker comparison
report = client.voice.forensics(
    primary_audio=open("interview_recording.mp3", "rb"),
    reference_audio=open("known_speaker_sample.wav", "rb"),
)

print(f"Risk Level: {report.overall_risk_level}")
print(f"Authenticity Score: {report.overall_score:.1%}")
print(f"Same Speaker: {report.speaker_comparison.is_same_speaker}")

for rec in report.recommendations:
    print(f"  - {rec}")

typescript

import { FotoHub } from "fotohub";

const client = new FotoHub({ apiKey: "fh_live_..." });

const report = await client.voice.forensics({
  primaryAudio: fs.createReadStream("interview_recording.mp3"),
  referenceAudio: fs.createReadStream("known_speaker_sample.wav"),
});

console.log(`Risk: ${report.overallRiskLevel}`);
console.log(`Score: ${(report.overallScore * 100).toFixed(1)}%`);
report.recommendations.forEach((r) => console.log(`  - ${r}`));

bash

curl -X POST "https://apis.fotohub.app/v1/ai/voice/forensics" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "primary_audio=@interview_recording.mp3" \
  -F "reference_audio=@known_speaker_sample.wav"

Supported Languages

Voice cloning and TTS generation support 23 languages with zero-shot capability — no language-specific training required for cloned voices.

Code	Language	Code	Language
`ar`	Arabic	`ms`	Malay
`da`	Danish	`nl`	Dutch
`de`	German	`no`	Norwegian
`el`	Greek	`pl`	Polish
`en`	English	`pt`	Portuguese
`es`	Spanish	`ru`	Russian
`fi`	Finnish	`sv`	Swedish
`fr`	French	`sw`	Swahili
`he`	Hebrew	`tr`	Turkish
`hi`	Hindi	`zh`	Chinese (Mandarin)
`it`	Italian
`ja`	Japanese
`ko`	Korean

Cross-lingual Cloning

A voice cloned from an English sample can generate speech in any of the 23 supported languages while maintaining the speaker's vocal characteristics. The system preserves timbre, pitch, and speaking patterns across languages.

Pricing

Operation	Credits	Description
Voice Clone	2	Create a new voice from reference audio
Clone Preview	Free	Test cloned voice (included with clone)
Emotional TTS	1	Generate speech with emotion control
Expressive TTS	1	Tag-based expressive generation
Standard TTS	1	Basic text-to-speech with cloned voice
Voice Comparison	1	Compare two voice samples
Deepfake Detection	2	Analyze audio for AI generation markers
Voice Forensics (full report)	3	Comprehensive forensic analysis

Credit Refunds

If a generation fails due to a system error (not invalid input), credits are automatically refunded to your account. Failed requests due to invalid audio format or empty text are not charged.

Best Practices

Reference Audio for Cloning

Requirement	Recommendation
Duration	10-30 seconds (20s optimal). Shorter clips reduce quality.
Format	WAV or FLAC preferred. MP3 acceptable but introduces compression artifacts.
Sample Rate	16 kHz minimum, 44.1 kHz recommended.
Environment	Quiet room, no background music or other speakers.
Content	Natural conversational speech. Avoid monotone reading.
File Size	Max 50 MB per file.
Speakers	Single speaker only. Multi-speaker audio will produce poor results.

Expressive Generation Tips

Start with moderate exaggeration (0.5-0.7) and increase gradually
Use cfg_weight to control how closely output follows the voice reference (higher = more controlled)
Combine emotion tags with pauses for natural-sounding transitions
Keep individual text segments under 300 characters for best quality
Use crossfade_ms: 50 (default) for smooth segment transitions
Emotion tags persist until the next tag — no need to repeat for consecutive sentences

Deepfake Detection Accuracy

Best results with 10-30 second audio clips
The model analyzes the middle 30 seconds of longer files
Higher accuracy on speech than music or environmental audio
Accuracy may be reduced on heavily compressed audio (low bitrate MP3)
For critical decisions, combine with the full forensics report

Error Handling

All endpoints return standard FOTOhub error responses:

json

{
  "detail": "Error description"
}

HTTP Code	Meaning
400	Invalid input (file too small, unsupported format, empty text)
401	Missing or invalid authentication token
402	Insufficient credits
413	File too large (exceeds 50 MB for clone, 200 MB for analysis)
502	GPU service temporarily unavailable
503	Model loading in progress (retry after 10-30 seconds)

Rate Limits

Endpoint	Limit
Voice Clone	10 requests/minute
Emotional/Expressive TTS	30 requests/minute
Voice Comparison	20 requests/minute
Deepfake Detection	15 requests/minute
Voice Forensics	5 requests/minute

Exceeding rate limits returns HTTP 429 with a Retry-After header indicating seconds until the next request window.

Accepted Audio Formats

All voice endpoints accept the following audio formats:

Format	Extensions	MIME Types
WAV	`.wav`	`audio/wav`, `audio/x-wav`
MP3	`.mp3`	`audio/mpeg`, `audio/mp3`
FLAC	`.flac`	`audio/flac`
AAC	`.aac`, `.m4a`	`audio/aac`, `audio/mp4`, `audio/x-m4a`
OGG	`.ogg`	`audio/ogg`
WebM	`.webm`	`audio/webm`

Voice Cloning & Verification ​

Voice Cloning Workflow ​

Endpoints ​

Create Voice Clone ​

Parameters ​

Response ​

Example ​

Preview Cloned Voice ​

Parameters ​

Response ​

Example ​

List Cloned Voices ​

Response ​

Example ​

Delete Cloned Voice ​

Response ​

Example ​

Emotional TTS ​

Emotion Dimensions ​

Parameters ​

Response ​

Example ​

Expressive TTS ​

Supported Tags ​

Available Emotions ​

Available Sound Effects ​

Parameters ​

Example ​

Voice Comparison ​

Parameters ​

Response ​

Example ​

Deepfake Detection ​

Parameters ​

Response ​

Example ​

Voice Forensics ​

Parameters ​

Response (Full Report) ​

Forensic Metrics Explained ​

Example ​

Supported Languages ​

Pricing ​

Best Practices ​

Reference Audio for Cloning ​

Expressive Generation Tips ​

Deepfake Detection Accuracy ​

Error Handling ​

Rate Limits ​

Accepted Audio Formats ​

Voice Cloning & Verification

Voice Cloning Workflow

Endpoints

Create Voice Clone

Parameters

Response

Example

Preview Cloned Voice

Parameters

Response

Example

List Cloned Voices

Response

Example

Delete Cloned Voice

Response

Example

Emotional TTS

Emotion Dimensions

Parameters

Response

Example

Expressive TTS

Supported Tags

Available Emotions

Available Sound Effects

Parameters

Example

Voice Comparison

Parameters

Response

Example

Deepfake Detection

Parameters

Response

Example

Voice Forensics

Parameters

Response (Full Report)

Forensic Metrics Explained

Example

Supported Languages

Pricing

Best Practices

Reference Audio for Cloning

Expressive Generation Tips

Deepfake Detection Accuracy

Error Handling

Rate Limits

Accepted Audio Formats