Connect with us

Spread & Containment

Synthetic speech startup Murf lends a voice to content creators of all sizes 

Synthetic speech tech startup Murf gives a voice, literally, to content creators of all sizes. Murf, which now has a library of more than 120 human-parity…

Published

on

Synthetic speech tech startup Murf gives a voice, literally, to content creators of all sizes. Murf, which now has a library of more than 120 human-parity AI voices across 20 languages, announced today it has raised $10 million Series A funding led by Matrix partners. Participation came from returning investors Elevation Capital and several prominent angel investors like Ola founder Ankit Bhai; Disney Streaming SVP of product; Ashwini Asokan, the founder of Mad Street Dap; and Pushkar Mukewar, founder Drip Capital

Founded in October 2020 by IIT-Kharagpur school friends Sneha Roy, Ankur Edkie and Divyanshu Pandey, Murf’s previous funding announcement was a $1.5 million seed led by Elevation Capital and angel investors who helped them recruit talent, invest in product innovation and user acquisition. Murf says that since its seed round, it has grown 26X in ARR and synthesized more than one million voice over projects, in wide variety of speaking styles and tones.

Some examples of how Murf’s technology has been used include a tech entrepreneur and artist creating an entire film using AI art models, deep fake programs and AI Voices from Murf studio; an entertainment animation agency that created a TV series using a collection of Murf’s voices; authors creating fantasy fiction audiobooks with Murf’s AI voices; and a YouTube influencer who used Murf’s AI voice to create a rap video.

Murf’s founders

Edkie, the CEO of Murf, told TechCrunch that even though Murf’s founding team worked in different domains in the past, they all ran into the pain points of creating high quality voice overs. This included creating and updating product demos and recording radio and video ads. He added that the pandemic “provided a boost to multimedia creation and the demand for scalable audio content was growing rapidly.”

Murf’s clients have used it in a variety of way, including advertising, audiobooks, explainer videos and e-learning. Murf.ai, its SaaS platform, was developed to make it easier for clients to create high-quality natural sounding voiceovers for any commercial purpose. The company’s clients range in size from individual content creators to SMBS and enterprises, and work in sectors like education, corporate, healthcare, media and entertainment, marketing, advertising, podcasting, customer support and more. 

Edkie, told TechCrunch that content creators and marketing teams often record voiceovers themselves, or outsource the entire process, both of which are “cumbersome, expensive and time-consuming.” Murf, on the other hand, lets users generate “human-like” voiceovers without needing to buy recording equipment or hire a voice artist. 

The company also wants to remove limitations on what text-to-speech can do. “While TTS has been around for quite some time now, limitations in voice quality have restricted its usage. By leveraging recent advances in AI and deep learning, we are making it possible to create high-fidelity synthetic voices that mimim the natural prosody and pronunciation of human speech.”

Murf’s platform includes an AI-enabled SaaS tool that helps users to generate “human-like” voices, typically for use in videos or presentations, without having procure complex and costly recoding equipment or hiring a voice artist. Content creators can use online voice recording booth, where they can sample a wide array of speaking styles. Murf wants to bridge diversity gap in traditional text to speech platforms by including voices across accents, like African American, British, Australian and others. 

According to market reports used by Murf’s founders, the global text-speech market is expected to reach $7.06 billion by 2028, growing at a 14.6% CAG. Meanwhile, the voiceover and dubbing markets is predicated to generate a total of $8 billion annually by 2027.

Text-to-speech has been around for years, but quality limitations meant they were used primarily by voice assistants and chat bots. But recent developments in AI and deep learning now means it is possible to create synthetic voices that have the prosody and pronunciation of human speech. Murf’s AI engine is trained on hours of actual human speech and Murf Studios offer more than 120 human-variety AI voices, which can speak in 20 languages. Murf is also working towards bringing more diverse accents by partnering with voice actors to bring abroad voices like African American, British and Australian English.

Murf’s AI-powered text-to-speech can also learn from contextual information to return the right responses. The founders describe Murf as an “all-in-one-voice solution” that enables users to add images, videos and background music. It also has features for pronunciation using the international phonetic alphabet (IPA), voice customizations that change users’ pitch, pause, emphasis and speed.

Murf makes money through a subscription plan for its services. It came out of beta testing in January 2021, and over the last 18 months, has grown 22x in ARR and synthesized over one million voiceover project to date.

Edkie said that Murf’s main competitors are the large tech and cloud companies, like Google, Amazon, Polly and Microsoft, who have the leading text to speech platforms in the market. Murf sets itself apart with natural sounding AI voices that also support multiple accents and styles. 

“Going beyond a simple text to speech tool, our platform offers the ability for users to add images, videos, presentations, and to the voiceover, include background music and sync them altogether to create compelling content,” said Edkie. Murf’s AI-powered TTS can also learn from large amounts of contextual information to create contextual speech. For example, it has an in-built context awareness that can recognize common used entity formats like numbers, currencies, percentages, addresses, dates and times, reducing their randomness and bringing them closer to a predefined standard, Edkie added. 

In a prepared statement, Elevation Capital co-managing partner Mukul Arora said, “AI-driven, life-like voiceovers are the next frontier in the text to speech market. Murf, with their stellar founding team and unique IP, is perfectly poised to gain a leadership position in this space. Their execution prowess and tech-first focus is evident in the solid traction and growth that they’ve demonstrated so far. We are really excited to double down on our  partnership with Murf.”

Synthetic speech startup Murf lends a voice to content creators of all sizes  by Catherine Shu originally published on TechCrunch

Read More

Continue Reading

Government

Coronavirus dashboard for October 5: an autumn lull as COVID-19 evolves towards seasonal endemicity

  – by New Deal democratBack in August I highlighted some epidemiological work by Trevor Bedford about what endemic COVID is likely to look like, based…

Published

on

 

 - by New Deal democrat

Back in August I highlighted some epidemiological work by Trevor Bedford about what endemic COVID is likely to look like, based on the rate of mutations and the period of time that previous infection makes a recovered person resistant to re-infection. Here’s his graph:




He indicated that it “illustrate[s] a scenario where we end up in a regime of year-round variant-driven circulation with more circulation in the winter than summer, but not flu-like winter seasons and summer troughs.”

In other words, we could expect higher caseloads during regular seasonal waves, but unlike influenza, the virus would never entirely recede into the background during the “off” seasons.

That is what we are seeing so far this autumn.

Confirmed cases have continued to decline, presently just under 45,000/day, a little under 1/3rd of their recent summer peak in mid-June. Deaths have been hovering between 400 and 450/day, about in the middle of their 350-550 range since the beginning of this past spring:



The longer-term graph of each since the beginning of the pandemic shows that, at their present level cases are at their lowest point since summer 2020, with the exception of a brief period during September 2020, the May-July lull in 2021, and the springtime lull this year. Deaths since spring remain lower than at any point except the May-July lull of 2021:



Because so many cases are asymptomatic, or people confirm their cases via home testing but do not get confirmation by “official” tests, we know that the confirmed cases indicated above are lower than the “real” number. For that, here is the long-term look from Biobot, which measures COVID concentrations in wastewater:



The likelihood is that there are about 200,000 “actual” new cases each day at present. But even so, this level is below any time since Delta first hit in summer 2021, with the exception of last autumn and this spring’s lulls.

Hospitalizations show a similar pattern. They are currently down 50% since their summer peak, at about 25,000/day:



This is also below any point in the pandemic except for briefly during September 2020, the May-July 2021 low, and this past spring’s lull.

The CDC’s most recent update of variants shows that BA.5 is still dominant, causing about 81% of cases, while more recent offshoots of BA.2, BA.4, and BA.5 are causing the rest. BA’s share is down from 89% in late August:



But this does not mean that the other variants are surging, because cases have declined from roughly 90,000 to 45,000 during that time. Here’s how the math works out:

89% of 90k=80k (remaining variants cause 10k cases)
81% of 45k=36k (remaining variants cause 9k cases)

The batch of new variants have been dubbed the “Pentagon” by epidmiologist JP Weiland, and have caused a sharp increase in cases in several countries in Europe and elsewhere. Here’s what she thinks that means for the US:


But even she is not sure that any wave generated by the new variants will exceed summer’s BA.5 peak, let alone approach last winter’s horrible wave:



In summary, we have having an autumn lull as predicted by the seasonal model. There will probably be a winter wave, but the size of that wave is completely unknown, primarily due to the fact that probably 90%+ of the population has been vaccinated and/or previously infected, giving rise to at least some level of resistance - a disease on its way to seasonal endemicity.

Read More

Continue Reading

Government

Gonorrhea became more drug resistant while attention was on COVID-19 – a molecular biologist explains the sexually transmitted superbug

The US currently has only one antibiotic available to treat gonorrhea – and it’s becoming less effective.

Published

on

By

The _Neisseria gonorrhoeae_ bacterium causes gonorrhea by infecting mucous membranes. Design Cells/iStock Getty Images Plus via Getty Images

COVID-19 has rightfully dominated infectious disease news since 2020. However, that doesn’t mean other infectious diseases took a break. In fact, U.S. rates of infection by gonorrhea have risen during the pandemic.

Unlike COVID-19, which is a new virus, gonorrhea is an ancient disease. The first known reports of gonorrhea date from China in 2600 BC, and the disease has plagued humans ever since. Gonorrhea has long been one of the most commonly reported bacterial infections in the U.S.. It is caused by the bacterium Neisseria gonorrhoeae, which can infect mucous membranes in the genitals, rectum, throat and eyes.

Gonorrhea is typically transmitted by sexual contact. It is sometimes referred to as “the clap.”

Prior to the pandemic, there were around 1.6 million new gonorrhea infections each year. Over 50% of those cases involved strains of gonorrhea that had become unresponsive to treatment with at least one antibiotic.

In 2020, gonorrhea infections initially went down 30%, most likely due to pandemic lockdowns and social distancing. However, by the end of 2020 – the last year for which data from the Centers for Disease Control and Prevention is available – reported infections were up 10% from 2019.

It is unclear why infections went up even though some social distancing measures were still in place. But the CDC notes that reduced access to health care may have led to longer infections and more opportunity to spread the disease, and sexual activity may have increased when initial shelter-in-place orders were lifted.

As a molecular biologist, I have been studying bacteria and working to develop new antibiotics to treat drug-resistant infections for 20 years. Over that time, I’ve seen the problem of antibiotic resistance take on new urgency.

Gonorrhea, in particular, is a major public health concern, but there are concrete steps that people can take to prevent it from getting worse, and new antibiotics and vaccines may improve care in the future.

How to recognize gonorrhea

Around half of gonorrhea infections are asymptomatic and can only be detected through screening. Infected people without symptoms can unknowingly spread gonorrhea to others.

Typical early signs of symptomatic gonorrhea include a painful or burning sensation when peeing, vaginal or penal discharge, or anal itching, bleeding or discharge. Left untreated, gonorrhea can cause blindness and infertility. Antibiotic treatment can cure most cases of gonorrhea as long as the infection is susceptible to at least one antibiotic.

There is currently only one recommended treatment for gonorrhea in the U.S. – an antibiotic called ceftriaxone – because the bacteria have become resistant to other antibiotics that were formerly effective against it. Seven different families of antibiotics have been used to treat gonorrhea in the past, but many strains are now resistant to one or more of these drugs.

The CDC tracks the emergence and spread of drug-resistant gonorrhea strains.

Why gonorrhea is on the rise

A few factors have contributed to the increase in infections during the COVID-19 pandemic.

Early in the pandemic, most U.S. labs capable of testing for gonorrhea switched to testing for COVID-19. These labs have also been contending with the same shortages of staff and supplies that affect medical facilities across the country.

Many people have avoided clinics and hospitals during the pandemic, which has decreased opportunities to identify and treat gonorrhea infections before they spread. In fact, because of decreased screening over the past two and a half years, health care experts don’t know exactly how much antibiotic-resistant gonorrhea has spread.

Also, early in the pandemic, many doctors prescribed antibiotics to COVID-19 patients even though antibiotics do not work on viruses like SARS-CoV-2, the virus that causes COVID-19. Improper use of antibiotics can contribute to greater drug resistance, so it is reasonable to suspect that this has happened with gonorrhea.

Overuse of antibiotics

Even prior to the pandemic, resistance to antibiotic treatment for bacterial infections was a growing problem. In the U.S., antibiotic-resistant gonorrhea infections increased by over 70% from 2017-2019.

Neisseria gonorrhoeae is a specialist at picking up new genes from other pathogens and from “commensal,” or helpful, bacteria. These helpful bacteria can also become antibiotic-resistant, providing more opportunities for the gonorrhea bacterium to acquire resistant genes.

Strains resistant to ceftriaxone have been observed in other countries, including Japan, Thailand, Australia and the U.K., raising the possibility that some gonorrhea infections may soon be completely untreatable.

Steps toward prevention

Currently, changes in behavior are among the best ways to limit overall gonorrhea infections – particularly safer sexual behavior and condom use.

However, additional efforts are needed to delay or prevent an era of untreatable gonorrhea.

Scientists can create new antibiotics that are effective against resistant strains; however, decreased investment in this research and development over the past 30 years has slowed the introduction of new antibiotics to a trickle. No new drugs to treat gonorrhea have been introduced since 2019, although two are in the final stage of clinical trials.

Vaccination against gonorrhea isn’t possible presently, but it could be in the future. Vaccines effective against the meningitis bacterium, a close relative of gonorrhea, can sometimes also provide protection against gonorrhea. This suggests that a gonorrhea vaccine should be achievable.

The World Health Organization has begun an initiative to reduce gonorrhea worldwide by 90% before 2030. This initiative aims to promote safe sexual practices, increase access to high-quality health care for sexually transmitted diseases and expand testing so that asymptomatic infections can be treated before they spread. The initiative is also advocating for increased research into vaccines and new antibiotics to treat gonorrhea.

Setbacks in fighting drug-resistant gonorrhea during the COVID-19 pandemic make these actions even more urgent.

Kenneth Keiler receives funding from NIH.

Read More

Continue Reading

Spread & Containment

Measuring the Ampleness of Reserves

Over the past fifteen years, reserves in the banking system have grown from tens of billions of dollars to several trillion dollars. This extraordinary…

Published

on

Over the past fifteen years, reserves in the banking system have grown from tens of billions of dollars to several trillion dollars. This extraordinary rise poses a natural question: Are the rates paid in the market for reserves still sensitive to changes in the quantity of reserves when aggregate reserve holdings are so large? In today’s post, we answer this question by estimating the slope of the reserve demand curve from 2010 to 2022, when reserves ranged from $1 trillion to $4 trillion.

What Are Reserves? And Why Do They Matter?

Banks hold accounts at the Federal Reserve where they keep cash balances called “reserves.” Reserves meet banks’ various needs, including making payments to other financial institutions and meeting regulatory requirements. Over the past fifteen years, reserves have grown enormously, from tens of billions of dollars in 2007 to $3 trillion today. The chart below shows the evolution of reserves in the U.S. banking system as a share of banks’ total assets from January 2010 through September 2022. The supply of reserves depends importantly on the actions of the Federal Reserve, which can increase or decrease the quantity of reserves by changing its securities holdings, as it did in response to the global financial crisis and the COVID-19 crisis.

Reserves Have Ranged from 8 to 19 Percent of Bank Assets from 2010 to 2022

Sources: Federal Reserve Bank of New York; Federal Reserve Economic Data, FRED (“TLAACBW027SBOG”); authors’ calculations.

Why does the quantity of reserves matter? Because the “price” at which banks trade their reserve balances, which in turn depends importantly on the total amount of reserves in the system, is the federal funds rate, which is the interest rate targeted by the Federal Open Market Committee (FOMC) in the implementation of monetary policy. In 2022, the FOMC stated that “over time, the Committee intends to maintain securities holdings in amounts needed to implement monetary policy efficiently and effectively in its ample reserves regime.” In this ample reserves regime, the Federal Reserve controls short-term interest rates mainly through the setting of administered rates, rather than by adjusting the supply of reserves each day as it did prior to 2008 (as discussed in this post). In today’s post, we describe a method to measure the sensitivity of interest rates to changes in the quantity of reserves that can serve as a useful indicator of whether the level of reserves is ample.

The Demand for Reserves Informs Us about Rate Sensitivity to Reserve Shocks

To assess whether the level of reserves is ample, one needs to first understand the demand for reserves. Banks borrow and lend in the market for reserves, typically overnight. The reserve demand curve describes the price at which these institutions are willing to trade their balances as a function of aggregate reserves. Its slope measures the price sensitivity to changes in the level of reserves. Importantly, banks earn interest on their reserve balances (IORB), set by the Federal Reserve. Because the IORB rate directly affects the willingness of banks to lend reserves, it is useful to describe the reserve demand curve in terms of the spread between the federal funds rate and the IORB rate. In addition, we control for the overall growth of the U.S. banking sector by specifying reserve demand in terms of the level of reserves relative to commercial banks’ assets.

There is a clear nonlinear downward-sloping relationship between prices and quantities of reserves, consistent with economic theory. The chart below plots the spread between the federal funds rate and the IORB against total reserves as a share of commercial banks’ total assets.  When reserves are very low, the demand curve has a steep negative slope, reflecting the willingness of borrowers to pay high rates because reserves are scarce. At the other extreme, when reserves are very high, the curve becomes flat because banks are awash with reserves and the supply is abundant. Between these two regions, an intermediate regime–that we refer to as “ample”–emerges, where the demand curve exhibits a modest downward slope. The color coding of the chart reflects the shifts in the reserve demand curve over time. In particular, the curve appears to have moved to the right and upward around 2015 and then moved upward after March 2020, at the onset of the COVID pandemic.

Reserve Demand Has Shifted over Time

Sources: Federal Reserve Bank of New York; Federal Reserve Economic Data, FRED (“TLAACBW027SBOG,” “IOER,” and “IORB”); authors’ calculations.

This chart highlights two of the main challenges in estimating the slope of the reserve demand curve. First, the curve is highly nonlinear, which means that a standard linear estimation approach is not appropriate. Second, various long-lasting changes in the regulation and supervision of banks, in their internal risk-management frameworks, and in the structure of the reserve market itself have resulted in shifts in the reserve demand curve. A third challenge is that the quantity of reserves may be endogenous to banks’ demand for them. Therefore, to properly measure the reserve demand curve, one must disentangle shocks to supply from those to demand. As we explain in detail in a recent paper, our estimation strategy addresses all three of these challenges.

Estimating the Slope of the Reserve Demand Curve

Our approach provides time-varying estimates of the price sensitivity of the demand for reserves that can be used to distinguish between periods in which reserves are relatively scarce, ample, or abundant. The chart below presents our daily estimates of the slope of the demand curve, as measured by the rate sensitivity to changes in reserves. Although we do not have a precise criterion for when reserves are scarce versus ample, during two episodes in our sample, the estimated rate sensitivity is well away from zero. The first episode occurs early in our sample, in 2010, and the second emerges almost ten years later, in mid-2019. In two other periods—during 2013-2017 and from mid-2020 through early September 2022—the estimated slope is very close to zero, indicating an abundance of reserves. The remaining periods are characterized by a modest negative slope of the reserve demand curve, consistent with ample (but short of abundant) reserves. The overall pattern of these estimates is robust to changes in the model specification, such as including spillovers from the repo and Treasury markets or measuring reserves as a share of gross domestic product or bank deposits (instead of as a share of banks’ assets).

Rate Sensitivity Changed over Time, Following the Path of Reserves

Sources: Federal Reserve Bank of New York; Federal Reserve Economic Data, FRED (“TLAACBW027SBOG,” “IOER,” and “IORB”); authors’ calculations.

Interest Rate Spreads Alone Are Not Reliable Indicators of Reserve Scarcity

As we discuss in our paper, the time variation in the estimated price sensitivity in the demand for reserves is based on observations of small movements along the demand curve due to exogenous supply shocks. The location of the curve itself, however, also changes over time. That is, there is not a constant relationship between the level of reserves and the slope of the reserve demand curve.  

In our paper, we find evidence of both horizontal and vertical shifts in the reserve demand curve, with vertical upward shifts being particularly important since 2015. This finding implies that the level of the federal funds-IORB spread may not be a reliable summary statistic for the sensitivity of interest rates to reserve shocks, and that estimates of the price sensitivity in the demand for reserves provide additional useful information.

In summary, we have developed a method to estimate the time-varying interest rate sensitivity of the demand for reserves that accounts for the nonlinear nature of reserve demand and allows for structural shifts over time. A key advantage of our methodology is that it provides a flexible and readily implementable approach that can be used to monitor the market for reserves in real time, allowing one to assess the “ampleness” of the reserve supply as market conditions evolve.

Gara Afonso is the head of Banking Studies in the Federal Reserve Bank of New York’s Research and Statistics Group.

Gabriele La Spada is a financial research economist in Money and Payments Studies in the Federal Reserve Bank of New York’s Research and Statistics Group.   

John C. Williams is the president and chief executive officer of the Federal Reserve Bank of New York.  

How to cite this post:
Gara Afonso, Gabriele La Spada, and John C. Williams, “Measuring the Ampleness of Reserves,” Federal Reserve Bank of New York Liberty Street Economics, October 5, 2022, https://libertystreeteconomics.newyorkfed.org/2022/10/measuring-the-ampleness-of-reserves/.


Disclaimer
The views expressed in this post are those of the author(s) and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. Any errors or omissions are the responsibility of the author(s).

Read More

Continue Reading

Trending