Bilingual (Hebrew-English) Audio Project: Agent–Customer Calls, Dual-Channel, Remote Version 2

PROJECT OVERVIEW We are recruiting bilingual speakers to participate in a contact-center style audio recording project. The work involves recording natural, realistic agent–customer telephone conversations in pairs, with each speaker assigned a dedicated language role:  Agent — speaks the secondary language only (typically English) Customer — speaks the primary language only (their native language)   The output is a single dual-channel WAV file per conversation, with the agent and customer captured on separate audio channels. The recordings are used to train and evaluate Automatic Speech Recognition (ASR) systems for multilingual contact centers.  This is a fixed-price, milestone-based engagement. Quality, naturalness, and authenticity are the primary success criteria.  SCOPE OF WORK The contractor's responsibilities are limited to two activities:  1. Script Correction — We provide an initial conversation script per session. The contractor reviews and refines it for naturalness, cultural fit, domain accuracy, and authentic conversational flow before recording.  2. Audio Recording — The contractor records the corrected, approved script as a natural agent–customer phone conversation, captured as a dual-channel WAV file at the assigned native sample rate.  All other workflow components (transcription, annotation, transliteration, post-processing) are handled internally and are NOT part of this engagement.  RECORDING REQUIREMENTS - File format: WAV (uncompressed). MP3, M4A, OGG, AAC and other compressed formats are not accepted. - Channels: Dual-channel. Channel 1 = Agent, Channel 2 = Customer. - File output: ONE single dual-channel WAV file per conversation (not two mono files). - Sample rates: 8 kHz, 16 kHz, AND 48 kHz — recorded NATIVELY at each rate per assignment. - Downsampling and upsampling are STRICTLY NOT ALLOWED. Files must be captured natively at the assigned rate. - Bit depth: 16-bit minimum. - Signal-to-Noise Ratio: minimum 25 dB. - Duration per conversation: 5–30 minutes (target ~10 minutes of active speech). - Recording type: One conversation per session, with clear channel separation between agent and customer. CODE-SWITCHING FORMAT (TYPE 1: BILINGUAL AGENT–CUSTOMER INTERACTION) - Agent speaks 100% in the secondary language (e.g., English) for the entire conversation. - Customer speaks 100% in the primary language (their native language) for the entire conversation. - Neither speaker switches into the other's language. The bilingual character of the recording comes from the two speakers using different languages, not from one speaker mixing both. - The result is a 50/50 language balance at the conversation level, achieved through role assignment. CONVERSATION STYLE  - Multi-turn, real-world contact-center scenarios (banking, airline, healthcare, insurance, finance, IT, telecom, e-commerce, technical support, billing, account management). - Context-driven dialogue: clarifications, account verification, technical terminology, problem-solving, resolutions. - Natural tone with realistic pauses, brief overlaps, fillers, backchannels, and conversational rhythm. - Speakers must understand the script and speak naturally — NOT read it aloud verbatim. - Tone should match a real customer-service phone call: professional but human. No theatrical, dramatic, exaggerated, or voice-acted delivery. - AI-generated, synthetic, voice-cloned, or text-to-speech audio is strictly prohibited.  SPEAKER REQUIREMENTS - Native speaker of the assigned primary language (for the Customer role). - Strong, professional-level proficiency in the secondary language, typically English (for the Agent role). Native speakers preferred; highly proficient L2 speakers acceptable. - Clear pronunciation, consistent volume, and steady delivery. - Each speaker can record up to 30 minutes of cumulative audio (~6 conversations maximum). - Ability to maintain clarity, tone, character, and consistency throughout each session. - Access to a quiet recording environment and quality recording equipment (see below).   EQUIPMENT & ENVIRONMENT - Wired headset microphone or quality USB/XLR microphone for each speaker. Built-in laptop microphones are NOT acceptable. - Recording software capable of capturing dual-channel WAV files at the specified native sample rates (8, 16, and 48 kHz). - Quiet indoor recording environment with no background noise (no fans, air conditioners, traffic, TV, music, doorbells, notifications, or other voices audible). - No echo, distortion, clipping, or audible hum. - Phones and other notification devices on silent (not vibrate) for the duration of recording. - If recording remotely with a partner, the recording setup must preserve channel separation and audio quality on both sides — discuss the setup with us before starting.   QUALITY STANDARDS - Minimum 25 dB signal-to-noise ratio. - No echo, distortion, clipping, background noise, or cross-channel bleed. - Consistent volume and clarity throughout the recording. - Natural, human delivery — no robotic, monotone, or scripted-reading tone. - Channel separation: agent must be cleanly on Channel 1, customer cleanly on Channel 2, with minimal bleed. - Maximum 3–5 seconds of silence at the start and at the end. - Natural pauses (1–5 seconds) within the conversation are acceptable. Long silent stretches that pad duration are not.   QUALITY ASSURANCE & REWORK POLICY - All deliverables undergo multi-level quality review covering audio quality, channel separation, naturalness, language adherence, and SOP compliance. - Acceptance threshold: 95% pass rate or higher on reviewed samples. - Rejected files must be re-recorded at no additional cost within the same milestone window. - Rework must be completed within 5 business days of rejection notice unless otherwise agreed. - A pattern of repeated rejections (more than two on a single recording, or rejections across multiple recordings) may result in termination of the contract without payment for unaccepted deliverables. PAYMENT STRUCTURE - Fixed-price contract. - Milestone-based payments tied to batch deliveries. - Payment is released only after final quality approval of the deliverables in that milestone. - Final payout is calculated on the basis of approved, error-free audio duration. Rejected or unaccepted recordings do not count toward payable hours. - No payment for AI-generated or synthetic audio (automatic forfeiture if detected). - All payments are processed through Upwork only. No off-platform payments. TERMS & CONDITIONS CONFIDENTIALITY & NON-DISCLOSURE - All scripts, recordings, instructions, and project communications are confidential. - The contractor must NOT share, repost, repurpose, publicly discuss, post on social media, include in portfolios, or otherwise disclose any project content, including the existence of the project itself, the end client, or the specific scenarios used. INTELLECTUAL PROPERTY & OWNERSHIP - All deliverables — including audio recordings, corrected scripts, voice content, and any associated materials — are the exclusive property of the client upon delivery. - The contractor irrevocably assigns all rights, title, and interest in the deliverables to the client, including the rights to reproduce, modify, distribute, train AI/ML models on, sublicense, and create derivative works from the recordings. - The contractor grants a perpetual, worldwide, royalty-free license for the use of their voice and likeness in the recordings for the project's stated purposes (ASR/AI model training and evaluation, and any related commercial uses by the client). - The contractor warrants that all submitted content is original, does not infringe any third-party rights, and does not include any copyrighted material, real personal data, or material the contractor does not have the right to record. AUTHENTICITY & ANTI-FRAUD - All recordings must be of real human voices, recorded live by the assigned speakers. - AI-generated speech, synthetic voices, voice cloning, text-to-speech (TTS) output, deepfaked audio, and any form of artificial speech generation are strictly prohibited. - Reusing, duplicating, or splicing previously recorded content across deliverables is prohibited. - We reserve the right to run automated and manual authenticity checks on all deliverables. Any audio identified as synthetic or fraudulent will be rejected and the engagement may be terminated immediately. - Detected fraud results in: (a) forfeiture of payment for affected deliverables and the entire milestone, (b) immediate contract termination, and (c) reporting to Upwork for Terms of Service violation. DATA PROTECTION & PRIVACY - The contractor must NOT use any real personal data (real names, real account numbers, real phone numbers, real addresses, real medical or financial information) in recordings. All such details in the script are fictional placeholders. - If real PII inadvertently appears in a recording, the contractor must flag it immediately so the affected segment can be re-recorded. - All local copies of scripts and audio must be deleted within 7 days of confirmed upload. - Files must only be transferred via the secure folder/link we provide. Email, public file-sharing, and messaging apps are not acceptable. - Each speaker must sign a Consent & Data Usage Agreement before participating. TIMELINE & DELIVERY DISCIPLINE - Each milestone has a delivery deadline communicated at milestone start. - Delays must be communicated proactively, in writing, with at least 48 hours' notice. - Repeated missed deadlines without prior communication may result in milestone forfeiture or contract termination. - Contractor is responsible for ensuring all submitted files use the correct naming convention and pass technical specs before submission. REVISIONS & FINAL APPROVAL - Each deliverable allows up to two rounds of revision per recording at no additional cost. - If a deliverable still does not meet quality standards after two rounds of revision, it will be marked unacceptable and excluded from payable hours. - Final approval is at our sole discretion based on the quality criteria stated in this post and the SOP. TERMINATION - We may terminate the engagement at any time for: any violation, submission of synthetic/AI-generated audio, repeated quality failures, missed deadlines without communication, or any breach of these Terms & Conditions. - Upon termination, only deliverables that have already been accepted will be paid for. Pending or rejected deliverables will not be compensated. - All confidentiality, data protection, and IP assignment obligations survive termination. LIABILITY & INDEMNIFICATION - Total liability under this engagement is limited to the agreed contract value for accepted deliverables. HOW TO APPLY Please include the following in your application: 1. Your native language(s) and your fluency level in English (or the secondary language you can perform as the Agent role). 2. Whether you can provide your own recording partner (a second speaker for the opposite role), or whether you would prefer we pair you with one. 3. A brief description of your recording setup: microphone(s) used, recording software, room/environment, and whether you can record at native 8 kHz, 16 kHz, and 48 kHz. 4. A short voice sample (30–60 seconds) demonstrating your speaking voice in your assigned role's language. A clean phone-call-style read is fine. 5. Confirmation that you have read and accept the Recording Requirements, Quality Standards, Payment Structure, and Terms & Conditions stated above. 6. Your hourly rate or per-conversation rate (in USD), and your weekly availability. Applications missing any of the above items will not be reviewed. We are looking for reliable, detail-oriented contributors who take audio quality seriously. Successful contractors will be invited to additional batches as the project scales across more language pairs.