Decoding Trader Speak: How Symphony’s Cloud9 is Revolutionizing Voice Trading Intelligence

Try Cloud9 on your desk

It is Monday at 9:05 am. As a commodity trader, you dial to reach out to your counterparty to place an important order. You hit the button and yell into the microphone, trying to overcome the bustling noise on the trading desk:

You: Okay, guys, Dec Jan CS roll, I’ll be 45 bid guys.
Counterparty: Okay.

You glance at your monitor. The voice-to-text transcription-enabled blotter picks up your voice but displays:

You: And these gents see us we’re all, I’ll be 45 big guys.
Counterparty: OK.

Once again, the trade details are inaccurate. You will need to pause and amend them manually to ensure proper compliance, or risk needing to listen to the hundreds of calls you place per day to recall trade details in the event that the middle office contacts you.

You face this situation multiple times a day.

Voice trading happens in seconds across scattered conversations, causing accurate trade data to vanish among a sea of inaccurate automated output. To enable reliable automated workflows, accurate transcription of trade parameters, post market analysis and compliance details is essential.

This is where Symphony steps in with its Cloud9 transcription solution.

Symphony developed a solution that treats voice trading as a form of Domain-Specific Language (DSL), with terms of art akin to the medical or legal field. Trader speak relies on thousands of lingo variants, shorthand phrases and asset class terminology agreed upon across market counterparties. Given the high number of variants in lingo terminology, the key to the solution was to create a model trained with your voice, instead of attempting to capture the entire vocabulary of “trader speak”.

Symphony’s model for trader speak

Symphony’s current transcription model is the result of over nearly two years of testing with real trading transactions.

The engine works by fine-tuning a foundational model using audio data samples our customers provide, hence directly adapting to the voices and shorthand of the traders working at each company. This means that every customer has a dedicated model tuned exclusively for their organization.

At execution time, the model relies on a pre-processor that takes advantage of how Symphony controls the audio ingestion from its Cloud9 system. Before feeding the audio to the transcription model, the pre-processor applies the following transformations to optimize audio quality and ensure high-fidelity output:

  • Noise reduction to reduce ambient sounds and noise from the trading floor.
  • Echo cancellation to clear up line distortion.
  • Voice activity detection to isolate speech segments from silence via an in-house Convolutional Neural Network (CNN) connected to a fully connected neural network model.
  • Audio channel separation to extract individual speaker lines.

How do we compare with the street?

To validate the strategy, Symphony compared its model against eight well known commercial transcription providers. The experiments were evaluated using two metrics:

  • Word Accuracy (WAcc): The percentage of spoken words transcribed correctly, accounting for wrong, missing or added words.
  • Semantic Accuracy: How well the transcribed text captured speaker intent. For example, the words “Okay” and “OK” in a sentence will have a very high semantic score, while the words “bid” and “bit” will have a lower score.
Performance MetricCommercial Models AverageSymphony Dedicated Model
Word Accuracy (WAcc) 65.23% 87.06%
Semantic Accuracy83.50% 93.70%

Overall, we are proud that our models outperform industry benchmarks. Most commercial speech models fail on the trading desk because they are trained for general consumer use cases like television captioning, corporate meetings or standard medical dictation. Voice trading is a specialized niche. By training dedicated models on actual institutional audio, Symphony drastically reduces transcription errors where accuracy matters most.

Tailored transcription enhanced by your data

Symphony offers structured deployment models designed to balance performance with strict enterprise data boundaries.

When institutions consent to providing internal audio archives, such as historical call logs from Q1 of the previous year, Symphony’s process pipeline handles dataset extraction and model adaption.

  • Complete data segregation: Dedicated models are fine-tuned from Symphony’s multi-firm trader speak knowledge base using your organization’s data, which is never pooled or shared for other firms’ models.
  • Staging and benchmark validation: Newly trained models deploy to a secure staging environment for evaluation against accuracy benchmarks.
  • Production deployment: Once metrics reach optimal performance, the custom engine transitions to production, available exclusively to your firm.

Firms preferring not to share internal audio to enhance their transcription accuracy can take advantage of Symphony’s robust base “trader speak” transcription model.

Integration with Cloud9’s compliance ecosystem

Operating natively within Symphony’s compliance architecture, Cloud9 manages customer usage data – including call audio, transcripts and metadata – in strict accordance with regional storage mandates, customer defined retention policies and customer preferences for data encryption. Transcriptions generated by Symphony models are accessible through interactive UI views or programmatic API distribution, leveraging event-driven triggers for real-time notification upon availability.

For firms that choose to record exclusively on premise, Symphony offers transcription as a standalone cloud service. This capability transcribes audio recorded locally with the Trader Speak transcription model without keeping any of the customer’s data on the cloud, thus making transcription available to on premise archival systems.

Going forward

Symphony is actively expanding beyond traditional compliance by building capabilities to convert raw trader voice streams into real-time, actionable market intelligence. By integrating Cloud9’s voice infrastructure with in-stream AI, real-time transcription and analytics, Symphony automates front-office trade capture, streamlines reconciliation and powers dynamic risk and performance dashboards. Transitioning from reactive voice archiving to active AI-driven intelligence will allow financial institutions to reduce operational risk, stay audit ready and maximize value from every trading interaction.

Try Cloud9 on your desk